博士（工学）, 大阪大学

Ph.D (Eng.), Osaka University

Materials Informatics

Machine Learning

Neural Network Model

Deep Learning

Image Processing

Medical Image Processing

Informatics, Soft computing

Informatics, Intelligent informatics

Apr. 2015
University of Electro Commnications, Professor, Japan

Mar. 2008 - Mar. 2015
University of Electro Commnications, Associate Professor, Japan

Oct. 2002 - Mar. 2008
Yamaguchi University, Associate Professor, Japan

Apr. 2001 - Sep. 2002
Nara Womens' University, Research Associate, Japan

Apr. 1994 - Mar. 2001
Osaka University, Faculty of Engineering Science, Research Associate, Japan

Apr. 1992 - Mar. 1994
Osaka University, Graduate School, Division of Engineering Science, Faculty of Biophysical Engineering, Japan

Apr. 1988 - Mar. 1992
Osaka University, Faculty of Engineering Science, Department of Biophysical Engineering, Japan

Apr. 1985 - Mar. 1988
Sendai-Dai-ni high school

Apr. 2023 - Present
Chair, IPSJ Transaction of Mathematical Modeling and its Applications, Society

Apr. 2020 - Mar. 2024
Goverment Board, Japan Neural Network Society (JNNS), Society

Jan. 2020 - Dec. 2023
Board of Government, Asia Pacific Neural Network Society, Society

Apr. 2013 - Mar. 2023
「数理モデル化と応用」副編集委員長, 情報処理学会, Society

Jan. 2022 - Dec. 2022
Finance Chair, International Conference on Neural Information Processing, Society

Apr. 2021
Member, IEEE Young Researcher Award Selection Comittee, Society

Apr. 2019 - Mar. 2021
Chair, IEEE Young Researcher Awards selection committee (Neuro computing), Society

Apr. 2018 - Mar. 2019
Member, IEEE Young Researcher Award Selection Comittee, Society

Apr. 2018 - Mar. 2019
Chair, Institute of Electronics, Information, and Communication Engineers, Technical Committee on Neurocomputing(NC), Society

Apr. 2017 - Mar. 2018
Vice-Chair, Institute of Electronics, Information, and Communication Engineers, Technical Committee on Neurocomputing(NC), Society

Apr. 2015 - Mar. 2018
Chair, Information Processing Society of Japan, Technical Committee of "Mathematics and Problem Solving (MPS)", Society

Jan. 2013 - Dec. 2015
Treasure, IEEE CIS Japan Chapter, Society

May 2012 - Apr. 2015
Goverment Board, Japan Neural Network Society (JNNS), Society

Apr. 2013 - Mar. 2015
Secretary, Information Processing Society of Japan, Technical Committee of "Mathematics and Problem Solving (MPS)", Society

Oct. 2013 - Sep. 2014
FIT2014 Exective Committee, Information Processing Society of Japan, Forum on Information Technology (FIT), Society

Oct. 2012 - Sep. 2013
FIT 2013 実行委員, 情報処理学会, Society

Apr. 2013
医用画像研究会連絡委員, 電子情報通信学会, Society

May 2012 - Mar. 2013
医用画像特別号編集委員, 電子情報通信学会, Society

Apr. 2008 - Mar. 2012
情報処理学会論文誌編集委員, 情報処理学会, Society

Apr. 2004 - Mar. 2012
MPS研究会副幹事 (Web担当), 情報処理学会, Society

Apr. 2002 - Mar. 2012
「数理モデル化と応用」編集委員, 情報処理学会, Society

May 2006 - May 2010
和文論文誌 D編集委員, 電子情報通信学会, Society

Apr. 2005 - Mar. 2007
中国支部学生会顧問, 電子情報通信学会, Society

Apr. 2003 - Mar. 2005
論文誌編集委員, 日本神経回路学会, Society

Apr. 2003
和文論文誌A 常任査読委員, 電子情報通信学会, Society

Apr. 2001
情報処理学会「数理モデル化と問題解決（MPS）研究会」連絡委員, 情報処理学会, Society

Apr. 1995 - Mar. 1997
ニューロコンピューティング研究会幹事補佐, 電子情報通信学会, Society

Mar. 2021
IEEE CIS Japan Chapter
Adversarial Training with Knowledge Distillation considering Intermediate Feature Representation in CNNs
IEEE Young Researcher Award, Hikaru Higuchi
Japan society

Dec. 2020
Japan Neural Network Society
VGGモデルの視覚野的解釈における解析の検討
日本神経回路学会最優秀研究賞, 寺本 陶冶;庄野 逸
Japan society

Mar. 2020
IEEE CIS Japan Chapter
スパースコーディングを用いた惑星表面画像のための圧縮方法の提案
IEEE Young Researcher Award, 上坂 佳史
Japan society

Mar. 2020
IEEE CIS Japan Chapter
Bolasso 特徴選択手法を用いたびまん性肺疾患陰影の分析
IEEE Young Researcher Award, 遠藤 瑛泰
Japan society

Jul. 2019
World academy of Science
Achievement award
International society

Mar. 2019
IEEE CIS Japan Chapter
SVCCAを用いた異なるデータセットで訓練されたDCNNの類似性測定
IEEE Young Researcher Award, 寺元 陶冶
Japan society

Mar. 2019
情報処理学会
2段階転移学習を用いた深層畳み込みニューラルネットによるびまん性肺疾患の識別と特徴表現の解析
山下記念賞, Aiga Suzuki
Japan society

Dec. 2018
情報処理学会数理モデル化と問題解決研究会
問題への適切性を考慮した畳み込みニューラルネットワークの初期値決定手法
MPS研究会ベストプレゼンテーション賞, 鈴木 藍雅;庄野 逸;坂無 英徳
Japan society

Dec. 2018
情報処理学会数理モデル化と問題解決研究会
ベイズ的変数選択に基づく分光スペクトル分解
MPS研究会ベストプレゼンテーション賞, 川島 貴大;庄野 逸
Japan society

Jul. 2018
情報処理学会数理モデル化と問題解決研究会
びまん性肺疾患診断における階層的特徴選択アプローチ
MPS研究会ベストプレゼンテーション賞, 遠藤 瑛泰;永田 賢二;庄野 逸
Japan society

Jul. 2018
World academy of Science
Achievement award
International society

Mar. 2018
IEEE CIS Japan Chapter
階層型確率的主成分分析モデルによるテクスチャの生成
IEEE Young Researcher Award, 鈴木藍雅
International society

Mar. 2018
情報処理学会数理モデル化と問題解決研究会
2段階転移学習を用いた深層畳み込みニューラルネットによるびまん性肺疾患の識別と特徴表現の解析
MPS研究会ベストプレゼンテーション賞, Aiga Suzuki;Hidenori Sakanashi;Shoji Kido;Hayaru Shouno
Japan society

Jul. 2017
World academy of Science
Achievement award
International society

Jul. 2016
情報処理学会
びまん性肺疾患識別におけるDeep Convolutional Neural Network特徴の解析
CS領域奨励研究賞, Satoshi Suzuki
Japan society

Jul. 2016
World academy of Science
Achievement award
International society

Mar. 2016
IEEE CIS Japan Chapter
Deep Convolutional Neural Networkを用いたびまん性肺疾患画像の解析
IEEE Young Researcher Award, 鈴木聡志
International society

Sep. 2015
情報処理学会数理モデル化と問題解決研究会
びまん性肺疾患識別におけるDeep Convolutional Neural Network特徴の解析
MPS研究会ベストプレゼンテーション賞, Satoshi Suzuki;Hayaru Shouno;Shoji Kido
Japan society

Jul. 2015
World academy of Science
Achievement award
International society

Sep. 2012
神経回路学会
相関トポグラフィック分析と自然画像への応用
日本神経回路学会大会奨励賞, 佐々木博昭
Japan society

Equivalent Circuit for Single/Three Phase Magnetic Coupling With Graph Neural Networks
Yusuke Yamakaji; Hayaru Shouno; Kunihiko Fukushima
IEEE Transactions on Power Electronics, Feb. 2025
Scientific journal

Model-Based Counterfactual Explanations Incorporating Feature Space Attributes for Tabular Data
Yuta Sumiya; Hayaru Shouno
Last, 2024 International Joint Conference on Neural Networks (IJCNN), IEEE, 31, 1-10, 30 Jun. 2024, Peer-reviwed
International conference proceedings

Bayesian inference method utilizing SESSA in quantitative layer structure estimation from XPS data
Atsushi Machida; Kenji Nagata; Ryo Murakami; Hiroshi Shinotsuka; Hayaru Shouno; Hideki Yoshikawa; Masato Okada
Journal of Electron Spectroscopy and Related Phenomena, Elsevier BV, 273, 147449-147449, Jun. 2024, Peer-reviwed
Scientific journal

Circuit2Graph: Circuits With Graph Neural Networks
Yusuke Yamakaji; Hayaru Shouno; Kunihiko Fukushima
IEEE Access, 12, 51818-51827, Apr. 2024, Peer-reviwed
Scientific journal, English

MAGRes-UNet: Improved Medical Image Segmentation Through a Deep Learning Paradigm of Multi-Attention Gated Residual U-Net
Tahir Hussain; Hayaru Shouno
Last, IEEE Access, 12, 40290-40310, Mar. 2024, Peer-reviwed
Scientific journal, English

Circuit2Graph: Diodes as Asymmetric Directional Nodes
Yusuke Yamakaji; Hayaru Shouno; Kunihiko Fukushima
IEEE Access, 2024
Scientific journal

Explainable Deep Learning Approach for Multi-Class Brain Magnetic Resonance Imaging Tumor Classification and Localization Using Gradient-Weighted Class Activation Mapping
Tahir Hussain; Hayaru Shouno
Last, Information, 30 Nov. 2023, Peer-reviwed
Scientific journal, English

Exploring the role of texture features in deep convolutional neural networks: Insights from Portilla-Simoncelli statistics
Yusuke Hamano; Shoko Nagasaka; Hayaru Shouno
Last, Neural Networks, 168, 300-312, Nov. 2023, Peer-reviwed, It is well-understood that the performance of Deep Convolutional Neural Networks (DCNNs) in image recognition tasks is influenced not only by shape but also by texture information. Despite this, understanding the internal representations of DCNNs remains a challenging task. This study employs a simplified version of the Portilla-Simoncelli Statistics, termed “minPS,” to explore how texture information is represented in a pre-trained VGG network. Using minPS features extracted from texture images, we perform a sparse regression on the activations across various channels in VGG layers. Our findings reveal that channels in the early to middle layers of the VGG network can be effectively described by minPS features. Additionally, we observe that the explanatory power of minPS sub-groups evolves as one ascends the network hierarchy. Specifically, sub-groups termed Linear Cross Scale (LCS) and Energy Cross Scale (ECS) exhibit weak explanatory power for VGG channels. To investigate the relationship further, we compare the original texture images with their synthesized counterparts, generated using VGG, in terms of minPS features. Our results indicate that the absence of certain minPS features suggests their non-utilization in VGG's internal representations.
Scientific journal, English

Sample structure prediction from measured XPS data using Bayesian estimation and SESSA simulator
Hiroshi Shinotsuka; Kenji Nagata; Malinda Siriwardana; Hideki Yoshikawa; Hayaru Shouno; Masato Okada
Journal of Electron Spectroscopy and Related Phenomena, Elsevier BV, 267, 147370-147370, Aug. 2023, Peer-reviwed
Scientific journal, English

Distorted image classification using neural activation pattern matching loss
Satoshi Suzuki; Shoichiro Takeda; Ryuichi Tanida; Yukihiro Bandoh; Hayaru Shouno
Last, Neural Networks, Elsevier BV, 167, 50-64, Aug. 2023, Peer-reviwed
Scientific journal, English

Correlation analysis with measurement conditions and peak structures in XPS spectral round-robin tests on MnO powder sample
Ryo Murakami; Yoshitomo Harada; Yutaka Sonobayashi; Hiroshi Oji; Hisao Makino; Hiromi Tanaka; Hideyuki Taguchi; Takanori Sakamoto; Haruka Morita; Akihiko Wakamori; Naoko Kibe; Shinsuke Nishida; Kenji Nagata; Hiroshi Shinotsuka; Hayaru Shouno; Hideki Yoshikawa
Journal of Electron Spectroscopy and Related Phenomena, Elsevier BV, 264, 147298-147298, Apr. 2023, Peer-reviwed
Scientific journal

Automatic estimation of unknown chemical components in a mixed material by XPS analysis using a genetic algorithm
Ryo Murakami; Hideki Yoshikawa; Kenji Nagata; Hiroshi Shinotsuka; Hiromi Tanaka; Takeshi Iizuka; Hayaru Shouno
Science and Technology of Advanced Materials: Methods, Talor & Francis Online, 2, 1, 91-105, 31 Dec. 2022, Peer-reviwed
Scientific journal, English

Fracture mode classification by texture analysis of fracture surface scanning electron microscope images
Akihiro Endo; Yoshiyuki Furuya; Kenji Nagata; Hideki Yoshikawa; Hayaru Shouno
Science and Technology of Advanced Materials: Methods, Talor & Francis Online, 2, 1, 129-138, 31 Dec. 2022, Peer-reviwed
Scientific journal, English

Inverse estimation of parameters for the magnetic domain via dynamics matching using visual-perceptive similarity
Ryo Murakami; Masaichiro Mizumaki; Ichiro Akai; Hayaru Shouno
Last, Science and Technology of Advanced Materials: Methods, Informa UK Limited, 2, 1, 139-152, 31 Dec. 2022, Peer-reviwed
Scientific journal, English

Adversarial Training with Knowledge Distillation Considering Intermediate Representations in CNNs
Hikaru Higuchi; Satoshi Suzuki; Hayaru Shouno
Last, Proc. Neural Information Processing. ICONIP2022 Communications in Computer and Information Science, Springer, IV, 683-691, 20 Nov. 2022, Peer-reviwed
International conference proceedings, English

A Lossless Audio Codec Based on Hierarchical Residual Prediction
Taiyo Mineo; Hayaru Shouno
Last, in Proceedings of Asia Pacific Signal Processing Association Annual Summit and Conference (APSIPA ASC), IEEE, to appear, 07 Nov. 2022, Peer-reviwed
International conference proceedings, English

Calculation of Spectral Similarity Independent of Measurement Equipment
Ryo Murakami; Hiroshi Shinotsuka; Kenji Nagata; Hideki Yoshikawa; Hayaru Shouno
Last, in Proceedings of Intl' Conference on Parallel Distributed Processing Techniques and its Applications, Nature Springer, to appear, 25 Jul. 2022, Peer-reviwed
International conference proceedings, English

Knowledge Transferred Fine-Tuning: Convolutional Neural Network Is Born Again With Anti-Aliasing Even in Data-Limited Situations
Satoshi Suzuki; Shoichiro Takeda; Naoki Makishima; Atsushi Ando; Ryo Masumura; Hayaru Shouno
IEEE Access, IEEE, 10, 68384-68396, 24 Jun. 2022, Peer-reviwed
Scientific journal, English

Improving sign-algorithm convergence rate using natural gradient for lossless audio compression
Taiyo Mineo; Hayaru Shouno
EURASIP Journal on Audio, Speech, and Music Processing, Springer, 2022, 12, 21 May 2022, Peer-reviwed
Scientific journal, English

Measuring Shift-invariance of Convolutional Neural Network with a Probability-incorporated Metric
Hikaru Higuchi; Satoshi Suzuki; Hayaru Shouno
Proc. Neural Information Processing. ICONIP2021 Communications in Computer and Information Science, Springer, 1516, 719-728, 02 Dec. 2021, Peer-reviwed
International conference proceedings, English

Prediction of metal temperature by microstructural features in creep exposed austenitic stainless steel with sparse modeling
Akihiro Endo; Kota Sawada; Kenji Nagata; Hideki Yoshikawa; Hayaru Shouno
Science and Technology of Advanced Materials: Methods, Talor & Francis Online, 1, 1, 225-233, 17 Nov. 2021, Peer-reviwed, This study proposes a framework to estimate the metal temperature from an optical micrograph of metals by using a machine learning approach. Specifically, 38 image statistical parameters such as area, contour, and circularity are calculated for the precipitate region determined through optical microscopy. Sparse modeling is then conducted to build a statistical model to estimate the Larson-Miller parameter (LMP), which is generally used in the evaluation of creep strength. This allows for the prediction of the metal temperature from the optical micrographs. The prediction performance of the proposed method is analyzed by applying it to KA-SUS304J1HTB (18Cr-9Ni-3Cu-Nb-N steel), reported in the NIMS Creep Data Sheets No. 56A and No. M-11. Consequently, temperature prediction is successfully achieved for unknown data with an error within ± 10°C.
Scientific journal, English

Determination of common peak structure from multiple X-ray photo-electron spectroscopy data sets
Ryo Murakami; Hayaru Shouno; Kenji Nagata; Hiroshi Shinotsuka; Hideki Yoshikawa
Science and Technology of Advanced Materials: Methods, Talor & Francis Online, 1, 1, 189-191, 03 Nov. 2021, Peer-reviwed, X-ray photo-electron spectroscopy (XPS) peak structure (i.e. peak parameters and the number of peaks) offers critical insights in chemical analysis of materials. Reference XPS spectral data are available for single-phases of compounds, as cited in various research papers and databases. Herein, we consider how individual peak structure varies among different reference spectra for the same single-phase of a compound. We developed a technique that automatically estimates common peak structures from multiple spectral data sets. Specifically, we developed a peak separation method that considers both common peak parameters and measurement-derived fluctuations. The proposed method can uniquely estimate the common peak structure of multiple XPS spectral data sets. For example, we applied the proposed approach to Ti 2p XPS results for TiO2 from 15 previous reports. In this way, we confirmed that estimated structure has high interpret-ability.
Scientific journal, English

Improving Convergence Rate of Sign Algorithm using Natural Gradient Method
Taiyo Mineo; Hayaru Shouno
The 29th in Proceedings of European Signal Processing Conference, IEEE, 51-55, 25 Aug. 2021, Peer-reviwed, In lossless audio compression, it is essential for predictive residuals to remain sparse when applying entropy codings. Hence, developing an accurate predictive method is crucial. The sign algorithm (SA) is a conventional method for minimizing the magnitude of residuals; however, it exhibits poor convergence performance compared with the least mean square (LMS) algorithm. To overcome the convergence performance degradation, we proposed novel adaptive algorithms based on a natural gradient: the natural gradient sign algorithm (NGSA) and normalized NGSA (NNGSA). We also propose an efficient update method for the natural gradient based on the AR(p) model. It requires O(p) multiply-add operations at every adaptation step. Through experiments conducted using toy data and real music data, we showed that the proposed algorithms achieve better convergence performance than the SA does. The NNGSA suggested having good compression ability in lossless audio coding.
International conference proceedings, English

Bayesian estimation for XPS spectral analysis at multiple core levels
Atsushi Machida; Kenji Nagata; Ryo Murakami; Hiroshi Shinotsuka; Hayaru Shouno; Hideki Yoshikawa; Masato Okada
Science and Technology of Advanced Materials: Methods, Talor & Francis Online, 1, 1, 123-133, 04 Aug. 2021, Peer-reviwed, X-ray photoelectron spectroscopy (XPS) is a widely used measurement technique in material surface analysis, but its analysis is subject to operator arbitrariness in the results. In a previous paper, a method based on genetic algorithms was proposed to estimate the composition ratios of compounds from XPS data using reference spectra and it was shown that it is possible to analyze them automatically from the reference spectra data. In this paper, we newly proposed a Bayesian spectral decomposition method based on the exchange Monte Carlo method and tested it on artificial data. This method provides a posterior distribution of the model parameters. This not only allows the estimation of compositional ratios for samples, but also allows statistical reliability assessment. In addition, we simulated an artificial data analysis to clarify the effect on the identification of compounds and the estimation of their compositional ratios by varying the signal-to-noise ratio of the data.
Scientific journal, English

NARU: Natural-gradient AutoRegressive Unlossy Audio Compressor
Taiyo Mineo; Hayaru Shouno
in Proceedings of Intl' Conference on Parallel Distributed Processing Techniques and its Applications, Nature Springer, to appear, 26 Jul. 2021, Peer-reviwed
International conference proceedings, English

Face Impression Classification in Cosmetic Counseling Using Deep Convolutional Neural Network
Masaharu Kurosawa; Hayaru Shouno
in Proceedings of Intl' Conference on Parallel Distributed Processing Techniques and its Applications, Nature Springer, to appear, 26 Jul. 2021, Peer-reviwed
International conference proceedings, English

Texture Analysis of Magnetic Domain Images Using Statistics Based on Human Visual Perception
Ryo Murakami; Masaichiro Mizumaki; Yusuke Hamano; Ichiro Akai; Hayaru Shouno
Journal of the Physical Society of Japan, 90, 4, 44705, 05 Mar. 2021, Peer-reviwed, In magnetic materials development, interpreting patterns of image data and estimating physical properties from image data are important. Specifically, magnetic domain images reflect the performance of magnetic materials. However, magnetic domain images are often evaluated qualitatively, i.e., they have a maze structure or an island structure. Therefore, this study quantitatively investigates the features describing the patterns of magnetic domain images, based on the Portilla-Simoncelli texture statistics (PSS). PSS is based on human visual perception, and is a strong tool to quantify texture structures. In the texture analysis of magnetic domain images, we primarily investigated the features describing texture structures, i.e., maze or island structures. In a secondary investigation, we estimated the physical properties using the texture statistics obtained from magnetic domain images. We determined the metrics of patterns from the magnetic domain images using PSS. Furthermore, we demonstrated that PSS can robustly estimate physical properties from magnetic domain images. Physical Society of Japan.
Scientific journal, English

Bayesian Dynamic Mode Decomposition with Variational Matrix Factorization
Takahiro Kawashima; Hayaru Shouno; Hideitsu Hino
In Proc. 35th AAAI Conference on Artificial Intelligence (AAAI-21), 35, 9, 8083-8091, 02 Feb. 2021, Peer-reviwed
International conference proceedings, English

Deep Feature Compression using Spatio-Temporal Arrangement toward Collaborative Intelligent World
Satoshi Suzuki; Shoichiro Takeda; Motohiro Takagi; Ryuichi Tanida; Hideaki Kimata; Hayaru Shouno
IEEE Transactions on Circuits and Systems for Video Technology, 2021, Collaborative Intelligence is a new paradigm that splits a deep neural network (DNN) into an edge and cloud for deploying a DNN-based image recognition application. In this paradigm, deep features, which are the outputs of the edge DNN, are compressed and transmitted to the cloud DNN. Because the deep features have a number of responses that are similar to each other, for efficient compression, previous methods spatially arrange and compress the deep features as an image to utilize the similarity as a spatial correlation. However, if the deep features are arranged in not only spatial but also temporal directions like those in a video, it may be possible to compress them more efficiently by increasing a temporal correlation. To explore this possibility, we propose a “spatio-temporal arrangement”. This method spatially arranges the deep features as images and temporally arranges them as a video with a novel ordering search algorithm. Our method effectively increases the spatial and temporal correlations hidden in the deep features and achieves high compression efficiency compared with the previous methods. Experimental results demonstrate the compression efficiency of our method is better than that of the previous methods (1.50% to 4.98% on BD-Rate evaluation in a lossy setting). Our analysis shows that our method effectively increases the correlation when the input is an image with rich edges and textures.
Scientific journal

Analysis of Texture Representation in Convolution Neural Network Using Wavelet Based Joint Statistics
Yusuke Hamano; Hayaru Shouno
Lecture Notes in Computer Science, Proc. of ICONIP, Springer International Publishing, 12532, 1, 126-136, 19 Nov. 2020, Peer-reviwed
International conference proceedings, English

Bayesian Sparse Covariance Structure Analysis for Correlated Count Data
Sho Ichigozaki; Takahiro Kawashima; Hayaru Shouno
Advances in Parallel & Distributed Processing, and Applications, Nature Springer, 781-791, 28 Jul. 2020, Peer-reviwed
International conference proceedings, English

Interpretation of ResNet by Visualization of Preferred Stimulus in Receptive Fields
Genta Kobayashi; Hayaru Shouno
Advances in Parallel & Distributed Processing, and Applications, Nature Springer, 769-779, 28 Jul. 2020, Peer-reviwed
International conference proceedings, English

Development of spectral decomposition based on Bayesian information criterion with estimation of confidence interval
Hiroshi Shinotsuka; Kenji Nagata; Hideki Yoshikawa; Yoh-ichi Mototake; Hayaru Shouno; Masato Okada
Science and Technology of Advanced Materials, Taylor & Francis, 21, 1, 402-419, 02 Jul. 2020, Peer-reviwed
Scientific journal, English

Development of multiple core-level XPS spectra decomposition method based on the Bayesian information criterion
Ryo Murakami; Hiromi Tanaka; Hiroshi Shinotsuka; Kenji Nagata; Hayaru Shouno; Hideki Yoshikawa
Journal of Electron Spectroscopy and Related Phenomena, Elsevier, 245, 147003-147003, 2020, Peer-reviwed
Scientific journal, English

B-DCGAN: Evaluation of Binarized DCGAN for FPGA
Hideo Terada; Hayaru Shouno
Lecture Notes in Computer Science, Proc. of ICONIP, Springer, 2, 55-64, 09 Dec. 2019, Peer-reviwed
International conference proceedings, English

Ordered Subset EM Algorithm for PET Image Reconstruction by use of Dictionary Learning and TV Regularization
Naohiro Okumura; Hayaru Shouno
日本医用画像工学会誌, 37, 5, 217-229, 29 Nov. 2019, Peer-reviwed
Scientific journal, Japanese

Deep Learning in Textural Medical Image Analysis
Aiga Suzuki; Hidenori Sakanashi; Shoji Kido; Hayaru Shouno
Intelligent Systems Reference Library, 171, 111-126, 18 Nov. 2019, © 2020, Springer Nature Switzerland AG. One of the characteristics of medical image analysis is that several medical images are not in the structure domain like natural images but in the texture domain. This chapter introduces a new transfer learning method, called “two-stage feature transfer,” to analyze textural medical images by deep convolutional neural networks. In the process of the two-stage feature transfer learning, the models are successively pre-trained with both natural image dataset and textural image dataset to get a better feature representation which cannot be derived from either of these datasets. Experimental results show that the two-stage feature transfer improves the generalization performance of the convolutional neural network on a textural lung CT pattern classification. To explain the mechanism of a transfer learning on convolutional neural networks, this chapter also shows analysis results of the obtained feature representations by an activation visualization method, and by measuring the frequency response of trained neural networks, in both qualitative and quantitative ways, respectively. These results demonstrate that such successive transfer learning enables networks to grasp both structural and textural visual features and be helpful to extracting good features from the textural medical images.
In book

Bolasso特徴選択手法を用いたびまん性肺疾患陰影の分析
遠藤 瑛泰; 永田 賢二; 木戸 尚治; 庄野 逸
電子情報通信学会技術研究報告(MEとバイオサイバネティックス), (一社)電子情報通信学会, 119, 224, 23-27, Oct. 2019, びまん性肺疾患は難病指定された疾患であり、異常陰影が肺X線CT画像上に現れる。異常陰影の様々なパターンは疾患や病変の性状を示すため、医師による異常陰影の特定は重要な役割を担う。異常陰影を識別するシステムを構築することにより、疾患の早期での発見や適切な治療へ繋がることが期待できる。しかし、これまでの先行研究において陰影クラス間での関係性などを考慮した識別システムの検討は十分に行われてはいない。そこで本研究では、様々な異常陰影に対して正常な陰影との一対一識別器の構築を行い、識別における正常な陰影との関係性を調査する。識別器の構築ではBolassoを用いた特徴選択を行い、異常陰影が持つ特徴表現を確認した。また、選択された特徴からクラス間の類似性を明らかにすることができた。(著者抄録)
Japanese

Fast Bayesian Restoration of Poisson Corrupted Images with INLA
Takahiro Kawashima; Hayaru Shouno
Proceedings of the 2019 International Conference on Parallel and Distributed Processing Techniques and Applications, CSREA Press, 1, 109-114, 26 Jul. 2019, Peer-reviwed
International conference proceedings, English

Spectral Deconvolution Based on Bayesian Variable Selection
Takahiro Kawashima; Hayaru Shouno
情報処理学会論文誌数理モデル化と応用, 情報処理学会, 12, 2, 34-43, 17 Jul. 2019, Peer-reviwed, In the field of spectroscopy, estimating the number of peaks and the shape and parameters of each peak from spectral data is a significant task. With respect to this task, Bayesian spectral deconvolution with replica exchange MCMC has been proposed and the effectiveness was shown. However, in Bayesian spectral deconvolution, we have to prepare multiple models and the computational time becomes large. Thus, in this study, we extended the previous model based on Bayesian variable selection, and we found that more efficient spectral deconvolution is possible for synthetic spectral data. In addition, we tried analyzing Raman spectrum of corundum and were able to extract valid peaks.
Scientific journal, Japanese

Deep Learning Employed in the Recognition of the Vector that Spreads Dengue, Chikungunya and Zika Viruses
Antonio Arista-Jalife; Alejandra Sanchez Oritz; Mariko Nakano; Henrik Tunnermann; Hector Perez-Meana; Hayaru Shouno
New Trends in Intelligent Software Methodologies, Tools, and Techniques, 108-120, 26 Sep. 2018, Peer-reviwed
International conference proceedings, English

Mosquito Larva Classification based on a Convolution Neural Network
Alejandra Sanchez Ortiz; Mariko Nakano; Henrik Tunnermann; Toya Teramoto; Hayaru Shouno
International Conference on Parallel and Distributed Processing Techniques and Applications, CSREA Press, 1, 320-325, 25 Jul. 2018, Peer-reviwed
International conference proceedings, English

Support Vector Machine Histogram: New Analysis and Architecture Design Method of Deep Convolutional Neural Network
Satoshi Suzuki; Hayaru Shouno
Neural Processing Letters, Springer New York LLC, 47, 3, 767-782, 01 Jun. 2018, Peer-reviwed, Deep convolutional neural network (DCNN) is a kind of hierarchical neural network models and attracts attention in recent years since it has shown high classification performance. DCNN can acquire the feature representation which is a parameter indicating the feature of the input by learning. However, its internal analysis and the design of the network architecture have many unclear points and it cannot be said that it has been sufficiently elucidated. We propose the novel DCNN analysis method “Support vector machine (SVM) histogram” as a prescription to deal with these problems. This is a method that examines the spatial distribution of DCNN extracted feature representation by using the decision boundary of linear SVM. We show that we can interpret DCNN hierarchical processing using this method. In addition, by using the result of SVM histogram, DCNN architecture design becomes possible. In this study, we designed the architecture of the application to large scale natural image dataset. In the result, we succeeded in showing higher accuracy than the original DCNN.
Scientific journal, English

Feature Representation Analysis of Deep Convolutional Neural Network using Two-stage Feature Transfer -An Application for Diffuse Lung Disease Classification-.
Aiga Suzuki; Hidenori Sakanashi; Shoji Kido; Hayaru Shouno
CoRR, 情報処理学会, abs/1810.06282, 11, 74-83, 2018, Peer-reviwed
Scientific journal, English

Simultaneous Estimation of Nongaussian Components and Their Correlation Structure
Hiroaki Sasaki; Michael U. Gutmann; Hayaru Shouno; Aapo Hyvarinen
NEURAL COMPUTATION, MIT PRESS, 29, 11, 2887-2924, Nov. 2017, Peer-reviwed, The statistical dependencies that independent component analysis (ICA) cannot remove often provide rich information beyond the linear independent components. It would thus be very useful to estimate the dependency structure from data. While such models have been proposed, they have usually concentrated on higher-order correlations such as energy (square) correlations. Yet linear correlations are a fundamental and informative form of dependency in many real data sets. Linear correlations are usually completely removed by ICA and related methods so they can only be analyzed by developing new methods that explicitly allow for linearly correlated components. In this article, we propose a probabilistic model of linear nongaussian components that are allowed to have both linear and energy correlations. The precision matrix of the linear components is assumed to be randomly generated by a higher-order process and explicitly parameterized by a parameter matrix. The estimation of the parameter matrix is shown to be particularly simple because using score-matching (Hyvarinen, 2005), the objective function is a quadratic form. Using simulations with artificial data, we demonstrate that the proposed method improves the identifiability of nongaussian components by simultaneously learning their correlation structure. Applications on simulated complex cells with natural image input, as well as spectrograms of natural audio data, show that the method finds new kinds of dependencies between the components.
Scientific journal, English

Analysis of Conventional Dropout and its Application to Group Dropout
Kazuyuki Hara; Daisuke Saitoh; Satoshi Suzuki; Takumi Kondou; Hayaru Shouno
情報処理学会論文誌数理モデル化と応用（TOM）, 10, 2, 25-32, 19 Jul. 2017, Deep learning is a state-of-the-art learning method that is used in fields such as visual object recognition and speech recognition. It uses very deep layers and a huge number of units and connections, so overfitting is a serious problem. The dropout method is used to address this problem. Dropout is a regularizer that neglects randomly selected inputs and hidden units during the learning process with probability q; after learning, the neglected inputs and hidden units are combined with the learned network to express the final output. Wager et al. pointed out that conventional dropout is an adaptive L2 regularizer, so we compared the learning behavior of conventional dropout with that of stochastic gradient descent with the L2 regularizer. We found that combining the neglected hidden units with the learned network can be regarded as ensemble learning, so we analyzed, on the basis of on-line learning, conventional dropout learning from the viewpoint of ensemble learning. Next we compared conventional dropout and ensemble learning from two additional viewpoints and confirmed that conventional dropout can be regarded as ensemble learning that divides a student network into two sub-networks. On the basis of this finding, we developed a novel dropout method that divides the network into more than two sub-networks. Computer simulation demonstrated that this method enhances the benefit of ensemble learning.Deep learning is a state-of-the-art learning method that is used in fields such as visual object recognition and speech recognition. It uses very deep layers and a huge number of units and connections, so overfitting is a serious problem. The dropout method is used to address this problem. Dropout is a regularizer that neglects randomly selected inputs and hidden units during the learning process with probability q; after learning, the neglected inputs and hidden units are combined with the learned network to express the final output. Wager et al. pointed out that conventional dropout is an adaptive L2 regularizer, so we compared the learning behavior of conventional dropout with that of stochastic gradient descent with the L2 regularizer. We found that combining the neglected hidden units with the learned network can be regarded as ensemble learning, so we analyzed, on the basis of on-line learning, conventional dropout learning from the viewpoint of ensemble learning. Next we compared conventional dropout and ensemble learning from two additional viewpoints and confirmed that conventional dropout can be regarded as ensemble learning that divides a student network into two sub-networks. On the basis of this finding, we developed a novel dropout method that divides the network into more than two sub-networks. Computer simulation demonstrated that this method enhances the benefit of ensemble learning.
English

Comparison of Feature Selection Method for Diffuse Lung Disease
Satoshi Ono; Makoto Koiwai; Hayaru Shouno; Shoji Kido
Proceedings of the 2017 International Conference on Parallel and Distributed Processing Techniques and Applications, CSREA Press, 1, 327-332, 17 Jul. 2017, Peer-reviwed
International conference proceedings, English

Generative Model of Textures Using Hierarchical Probabilistic Principal Component Analysis
Aiga Suzuki; Hayaru Shouno
Proceedings of the 2017 International Conference on Parallel and Distributed Processing Techniques and Applications, CSREA Press, 1, 333-338, 17 Jul. 2017, Peer-reviwed
International conference proceedings, English

A study on visual interpretation of network in network
Satoshi Suzuki; Hayaru Shouno
Proceedings of the International Joint Conference on Neural Networks, Institute of Electrical and Electronics Engineers Inc., 2017-, 903-910, 30 Jun. 2017, Peer-reviwed, In recent years, Deep convolutional neural networks (DCNNs) have shown excellent performance in the image recognition field. A DCNN is one of the types of multi-layer neural networks, which can automatically obtain feature representation from input data. The Neocognitron, proposed by Kunihiko Fukushima in the 1980s, is a prototype of a DCNN. It was inspired by the hierarchical structure in the mammalian primary visual cortex. On the other hand, a method called Network In Network (NIN) has recently been proposed. This is a network architecture that embeds some translational symmetric micro networks in a DCNN, and it has been experimentally clarified that with it higher classification accuracy is obtained than in a conventional DCNN. However, it cannot be said that NIN has been as sufficiently analyzed from a physiological point of view compared to DCNNs. We focused on the similarities between the processing of NIN, which accumulates the feature extraction filter of a DCNN, and the operation of a mammalian visual structure called an 'orientation continuity' which means preferred orientations of neighboring cells changes continuously, and pointed out the relationships between them. We also studied and pointed out the relevance of the neurophysiological knowledge and the process results obtained with high layer of NIN.
International conference proceedings, English

Feature Selection for Diffuse Lung Disease using Exchange Markov Chain Monte-Carlo Method
Makoto KOIWAI; Nodoka IIDA; Hayaru SHOUNO; Shoji KIDO
Proceedings of Parallel Distributed Processing Techniques and Applications, Athens, 1, 381-386, Jul. 2016, Peer-reviwed
International conference proceedings, English

Architecture Design of Deep Convolutional Neural Network for Diffuse Lung Disease Using Representation Separation Information
Satoshi SUZUKI; Nodoka IIDA; Hayaru SHOUNO; Shoji KIDO
Proceedings of Parallel Distributed Processing Techniques and Applications, Athens, 1, 387-393, Jul. 2016, Peer-reviwed
International conference proceedings, English

Analysis of Dropout Learning Regarded as Ensemble Learning
Kazuyuki Hara; Daisuke Saitoh; Hayaru Shouno
ARTIFICIAL NEURAL NETWORKS AND MACHINE LEARNING - ICANN 2016, PT II, SPRINGER INT PUBLISHING AG, 9887, 72-79, 2016, Peer-reviwed, Deep learning is the state-of-the-art in fields such as visual object recognition and speech recognition. This learning uses a large number of layers, huge number of units, and connections. Therefore, overfitting is a serious problem. To avoid this problem, dropout learning is proposed. Dropout learning neglects some inputs and hidden units in the learning process with a probability, p, and then, the neglected inputs and hidden units are combined with the learned network to express the final output. We find that the process of combining the neglected hidden units with the learned network can be regarded as ensemble learning, so we analyze dropout learning from this point of view.
International conference proceedings, English

Group Dropout Inspired by Ensemble Learning
Kazuyuki Hara; Daisuke Saitoh; Takumi Kondou; Satoshi Suzuki; Hayaru Shouno
NEURAL INFORMATION PROCESSING, ICONIP 2016, PT II, SPRINGER INT PUBLISHING AG, 9948, 66-73, 2016, Peer-reviwed, Deep learning is a state-of-the-art learning method that is used in fields such as visual object recognition and speech recognition. This learning uses a large number of layers and a huge number of units and connections, so overfitting occurs. Dropout learning is a kind of regularizer that neglects some inputs and hidden units in the learning process with a probability p; then, the neglected inputs and hidden units are combined with the learned network to express the final output. We compared dropout learning and ensemble learning from three viewpoints and found that dropout learning can be regarded as ensemble learning that divides the student network into two groups of hidden units. From this insight, we explored novel dropout learning that divides the student network into more than two groups of hidden units to enhance the benefit of ensemble learning.
International conference proceedings, English

An Architecture Design Method of Deep Convolutional Neural Network
Satoshi Suzuki; Hayaru Shouno
NEURAL INFORMATION PROCESSING, ICONIP 2016, PT III, SPRINGER INT PUBLISHING AG, 9949, 538-546, 2016, Peer-reviwed, Deep Convolutional Neural Network (DCNN) is a kind of multi layer neural network models. In these years, the DCNN is attracting the attention since it shows the state-of-the-arts performance in the image and speech recognition tasks. However, the design for the architecture of the DCNN has not so much discussed since we have not found effective guideline to construct. In this research, we focus on within-class variance of SVM histogram proposed in our previous work [8]. We try to apply it as a clue for modifying the architecture of a DCNN, and confirm the modified DCNN shows better performance than that of the original one.
International conference proceedings, English

Semi-supervised Based Learning for Idiopathic Interstitial Pneumonia on HRCT Images
Hayaru Shouno; Shoji Kido
Proceedings of Parallel Distributed Processing Techniques and Applications, 8, 27 Jul. 2015, Peer-reviwed
International conference proceedings, English

Quantitative Evaluation of Reconstructed Image with Filtered Back Projection Bayes Method
Nodoka Iida; Hayaru Shouno; Muneyuki Sakata; Yuichi Kimura
Proceedings of Parallel Distributed Processing Techniques and Applications, 8, 27 Jul. 2015, Peer-reviwed
International conference proceedings, English

Dark channel prior based blurred image restoration method using total variation and morphology
Yibing Li; Qiang Fu; Fang Ye; Hayaru Shouno
JOURNAL OF SYSTEMS ENGINEERING AND ELECTRONICS, SYSTEMS ENGINEERING & ELECTRONICS, EDITORIAL DEPT, 26, 2, 359-366, Apr. 2015, Peer-reviwed, The blurred image restoration method can dramatically highlight the image details and enhance the global contrast, which is of benefit to improvement of the visual effect during practical applications. This paper is based on the dark channel prior principle and aims at the prior information absent blurred image degradation situation. A lot of improvements have been made to estimate the transmission map of blurred images. Since the dark channel prior principle can effectively restore the blurred image at the cost of a large amount of computation, the total variation (TV) and image morphology transform (specifically top-hat transform and bottom-hat transform) have been introduced into the improved method. Compared with original transmission map estimation methods, the proposed method features both simplicity and accuracy. The estimated transmission map together with the element can restore the image. Simulation results show that this method could inhibit the ill-posed problem during image restoration, meanwhile it can greatly improve the image quality and definition.
Scientific journal, English

Bayesian Restoration for Poisson Corrupted Image using a Latent Variational Method with Gaussian MRF
Hayaru Shouno
情報処理学会誌 「数理モデル化と応用」, 情報処理学会, 8, 1, 62-71, 31 Mar. 2015, Peer-reviwed, We treat an image restoration problem with a Poisson noise channel using a Bayesian framework. The Poisson randomness might be appeared in observation of low contrast object in the field of imaging. The noise observation is often hard to treat in a theoretical analysis. In our formulation, we interpret the observation through the Poisson noise channel as a likelihood, and evaluate the bound of it with a Gaussian function using a latent variable method. We then introduce a Gaussian Markov random field (GMRF) as the prior for the Bayesian approach, and derive the posterior as a Gaussian distribution. The latent parameters in the likelihood and the hyperparameter in the GMRF prior could be treated as hidden parameters, so that, we propose an algorithm to infer them in the expectation maximization (EM) framework using loopy belief propagation (LBP). We confirm the ability of our algorithm in the computer simulation, and compare it with the results of other image restoration frameworks.
Scientific journal, English

Bayesian image restoration for poisson corrupted image using a latent variational method with gaussian MRF
Hayaru Shouno
IPSJ Online Transactions, Information Processing Society of Japan, 8, 2015, 15-24, 2015, Peer-reviwed, We treat an image restoration problem with a Poisson noise channel using a Bayesian framework. The Poisson randomness might be appeared in observation of low contrast object in the field of imaging. The noise observation is often hard to treat in a theoretical analysis. In our formulation, we interpret the observation through the Poisson noise channel as a likelihood, and evaluate the bound of it with a Gaussian function using a latent variable method. We then introduce a Gaussian Markov random field (GMRF) as the prior for the Bayesian approach, and derive the posterior as a Gaussian distribution. The latent parameters in the likelihood and the hyperparameter in the GMRF prior could be treated as hidden parameters, so that, we propose an algorithm to infer them in the expectation maximization (EM) framework using loopy belief propagation (LBP). We confirm the ability of our algorithm in the computer simulation, and compare it with the results of other image restoration frameworks.
Scientific journal, English

Deep Convolutional Network Neocognitron: Improved Interpolating-Vector
Kunihiko Fukushima; Hayaru Shouno
2015 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS (IJCNN), IEEE, 8, 2015, Peer-reviwed, The neocognitron is a multi-layered convolutional network that can be trained to recognize visual patterns robustly. In the intermediate layers of the neocognitron, local features are extracted from input patterns. In the highest (or deepest) layers of the network, the method of Interpolating-Vector is used for classifying patterns based on the features extracted by the intermediate layers. During the learning, several reference vectors for each class are created from a set of training vectors. To recognize an input vector, we measure distances (based on similarities) between the input vector and planes that are spanned by every trio of reference vectors of the same class. The class name of the nearest plane is taken as the result of classification. To reduce the computational cost, we propose to search the nearest plane, not among all possible combinations of three reference vectors, but only among trios that contain the nearest reference vector. For reducing the computational cost, it is also important to represent the large number of training vectors accurately with a compact set of reference vectors. To create a compact set of reference vectors, the learning is carried out in two steps. In the first step, reference vectors are just chosen from vectors in the training set. We start modifying reference vectors (namely, fine tuning of connections) from the second step after an enough number of reference vectors have been chosen. The effectiveness of the proposed method for recognizing hand-written digits is demonstrated by computer simulation.
International conference proceedings, English

Analysis of Function of Rectified Linear Unit Used in Deep learning
Kazuyuki Hara; Daisuke Saito; Hayaru Shouno
2015 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS (IJCNN), IEEE, 8, 2015, Peer-reviwed, Deep Learning is attracting much attention in object recognition and speech processing. A benefit of using the deep learning is that it provides automatic pre-training. Several proposed methods that include auto-encoder are being successfully used in various applications. Moreover, deep learning uses a multilayer network that consists of many layers, a huge number of units, and huge amount of data. Thus, executing deep learning requires heavy computation, so deep learning is usually utilized with parallel computation with many cores or many machines. Deep learning employs the gradient algorithm, however this traps the learning into the saddle point or local minima. To avoid this difficulty, a rectified linear unit (ReLU) is proposed to speed up the learning convergence. However, the reasons the convergence is speeded up are not well understood. In this paper, we analyze the ReLU by a using simpler network called the soft-committee machine and clarify the reason for the speedup. We also train the network in an on-line manner. The soft-committee machine provides a good test bed to analyze deep learning. The results provide some reasons for the speedup of the convergence of the deep learning.
International conference proceedings, English

A Transfer Learning Method with Deep Convolutional Neural Network for Diffuse Lung Disease Classification
Hayaru Shouno; Satoshi Suzuki; Shoji Kido
NEURAL INFORMATION PROCESSING, PT I, SPRINGER INT PUBLISHING AG, 9489, 199-207, 2015, Peer-reviwed, We introduce a deep convolutional neural network (DCNN) as feature extraction method in a computer aided diagnosis (CAD) system in order to support diagnosis of diffuse lung diseases (DLD) on high-resolution computed tomography (HRCT) images. DCNN is a kind of multi layer neural network which can automatically extract features expression from the input data, however, it requires large amount of training data. In the field of medical image analysis, the number of acquired data is sometimes insufficient to train the learning system. Overcoming the problem, we apply a kind of transfer learning method into the training of the DCNN. At first, we apply massive natural images, which we can easily collect, for the pre-training. After that, small number of the DLD HRCT image as the labeled data is applied for fine-tuning. We compare DCNNs with training of (i) DLD HRCT images only, (ii) natural images only, and (iii) DLD HRCT images + natural images, and show the result of the case (iii) would be better DCNN feature rather than those of others.
International conference proceedings, English

Acceleration of Poisson Corrupted Image Restoration with Loopy Belief Propagation
Hayaru Shouno
International Conference on Parallel and Distributed Processing Techniques and Applications, PDPTA, 1, 165-170, 21 Jul. 2014, Peer-reviwed
International conference proceedings, English

Estimating Dependency Structures for non-Gaussian Components with Linear and Energy Correlations
H. Sasaki; M. U. Gutmann; H. Shouno; A. Hyvärinen
Journal of Machine Learning Research, Journal of Machine Learning Research, 33, 868-876, Apr. 2014, Peer-reviwed
International conference proceedings, English

Distribution estimation of hyperparameters in Markov random field models
Yoshinori Nakanishi-Ohno; Kenji Nagata; Hayaru Shouno; Masato Okada
JOURNAL OF PHYSICS A-MATHEMATICAL AND THEORETICAL, IOP PUBLISHING LTD, 47, 4, Jan. 2014, Peer-reviwed, We developed a method of distribution estimation of hyperparameters in Markov random field (MRF) models. This study was motivated by the growing quantity of image data in natural sciences owing to recent advances in measurement techniques. MRF models are used to restore images in information science, and the hyperparameters of these models can be adjusted to improve restoration performance. The parameters appearing in data analysis represent physical quantities such as diffusion coefficients. Indeed, many frameworks of hyperparameter estimation have been proposed, but most are point estimation that is susceptible to stochastic fluctuations. Distribution estimation can be used to evaluate the confidence one has in point estimates of hyperparameters, in a similar way to physicists using error bars when they evaluate important physical quantities. We use a solvable MRF model to investigate the performance of distribution estimation in simulations.
Scientific journal, English

Dictionary-Based Image Denoising by Fused-Lasso Atom Selection
Ao Li; Hayaru Shouno
MATHEMATICAL PROBLEMS IN ENGINEERING, HINDAWI PUBLISHING CORP, 2014, 368602, 2014, Peer-reviwed, We proposed an efficient image denoising scheme by fused lasso with dictionary learning. The scheme has two important contributions. The first one is that we learned the patch-based adaptive dictionary by principal component analysis (PCA) with clustering the image into many subsets, which can better preserve the local geometric structure. The second one is that we coded the patches in each subset by fused lasso with the clustering learned dictionary and proposed an iterative Split Bregman to solve it rapidly. We present the capabilities with several experiments. The results show that the proposed scheme is competitive to some excellent denoising algorithms.
Scientific journal, English

Correlated topographic analysis: estimating an ordering of correlated components
Hiroaki Sasaki; Michael Gutmann; Hayaru Shouno; Aapo Hyvarinen
JMLR, JMLR.org, 25, 365-378, Sep. 2013, Peer-reviwed
International conference proceedings, English

Correlated topographic analysis: estimating an ordering of correlated components
Hiroaki Sasaki; Michael U. Gutmann; Hayaru Shouno; Aapo Hyvarinen
MACHINE LEARNING, SPRINGER, 92, 2-3, 285-317, Sep. 2013, Peer-reviwed, This paper describes a novel method, which we call correlated topographic analysis (CTA), to estimate non-Gaussian components and their ordering (topography). The method is inspired by a central motivation of recent variants of independent component analysis (ICA), namely, to make use of the residual statistical dependency which ICA cannot remove. We assume that components nearby on the topographic arrangement have both linear and energy correlations, while far-away components are statistically independent. We use these dependencies to fix the ordering of the components. We start by proposing the generative model for the components. Then, we derive an approximation of the likelihood based on the model. Furthermore, since gradient methods tend to get stuck in local optima, we propose a three-step optimization method which dramatically improves topographic estimation. Using simulated data, we show that CTA estimates an ordering of the components and generalizes a previous method in terms of topography estimation. Finally, to demonstrate that CTA is widely applicable, we learn topographic representations for three kinds of real data: natural images, outputs of simulated complex cells and text data.
Scientific journal, English

Poisson Observed Image Restoration using a Latent Variational Approximation with Gaussian MRF
H. Shouno; M. Okada
Proc on PDPTA2013, 1, 201-206, Jul. 2013, Peer-reviwed
International conference proceedings, English

Bayes Tomography Reconstruction using 4-Dimensional Markov Random Field Prior
M. Yamasaki; H. Shouno; M. Okada
電子情報学会論文誌 D, The Institute of Electronics, Information and Communication Engineers, J96-D, 4, 791-802, Apr. 2013, Peer-reviwed, 医療診断などで用いられる断層撮像法の画像再構成問題において,空間と時間からなる四次元の事前分布を導入したBayes画像推定法を提案し,画質による性能評価を行った.Bayes推定を行う場合,観測データの他に原画像がどのような画像であるべきかを記述する事前分布を画像モデルとして導入する必要がある.断層撮像法の被写体は,空間的には三次元の連続的な構造をもち時間的にも滑らかに変化していくと仮定し,時空間モデルとして四次元のMarkov確率場(MRF)状の関数を,可解モデルとなるように事前分布として導入した.事後分布は解析解として導出し,これを用いた画像再構成法に関して数値解を求めた.本提案手法と,(Filtered Back Projection)FBP法と呼ばれる従来手法,及び先行研究の二次元MRF状関数を事前分布として用いた手法との性能比較を行い,ノイズ除去の意味で良好な結果を得ることができた.
Scientific journal, Japanese

A hierarchical extension of the HOG model implemented in the convolutional-net for human detection
Yasuto Arakaki; Hayaru Shouno; Kazuyuki Takahashi; Takashi Morie
情報処理学会誌「数理モデル化と応用」, 情報処理学会, 5, 3, 54-62, Apr. 2013, Peer-reviwed
Scientific journal, English

Topographic analysis of correlated components
Sasaki, H.; Gutmann, M.U.; Shouno, H.; Hyvärinen, A.
Journal of Machine Learning Research, 25, 1-14, Sep. 2012, Peer-reviwed
Scientific journal, English

Deterministic Algorithm for Nonlinear Markov Random Field Model
Ohno, Yoshinori; Nagata, Kenji; Kuwatani, Tatsu; Shouno, Hayaru; Okada, Masato
Journal of the Physical Society of Japan, Physical Society of Japan, 81, 6, 064006-064006, Jun. 2012, Peer-reviwed
Scientific journal, English

A Hierarchical Extension of the HOG Model Implemented in the Convolution-net for Human Detection
Arakaki Yasuto; Shouno Hayaru; Takahashi Kazuyuki; Morie Takashi
IMT, Information and Media Technologies Editorial Board, 7, 4, 1480-1488, 2012, For the detection of generic objects in the field of image processing, histograms of orientation gradients (HOG) is discussed for these years. The performance of the classification system using HOG shows a good result. However, the performance of using HOG descriptor would be influenced by the detecting object size. In order to overcome this problem, we introduce a kind of hierarchy inspired from the convolution-net, which is a model of our visual processing system in the brain. The hierarchical HOG (H-HOG) integrates several scales of HOG descriptors in its architecture, and represents the input image as the combinatorial of more complex features rather than that of the orientation gradients. We investigate the H-HOG performance and compare with the conventional HOG. In the result, we obtain the better performance rather than the conventional HOG. Especially the size of representation dimension is much smaller than the conventional HOG without reducing the detecting performance.
English

A hierarchical extension of the HOG model implemented in the convolution-net for human detection
Yasuto Arakaki; Hayaru Shouno; Kazuyuki Takahashi; Takashi Morie
IPSJ Online Transactions, Information Processing Society of Japan, 5, 2012, 177-185, 2012, Peer-reviwed, For the detection of generic objects in the field of image processing, histograms of orientation gradients (HOG) is discussed for these years. The performance of the classification system using HOG shows a good result. However, the performance of using HOG descriptor would be influenced by the detecting object size. In order to overcome this problem, we introduce a kind of hierarchy inspired from the convolution-net, which is a model of our visual processing system in the brain. The hierarchical HOG (H-HOG) integrates several scales of HOG descriptors in its architecture, and represents the input image as the combinatorial of more complex features rather than that of the orientation gradients. We investigate the H-HOG performance and compare with the conventional HOG. In the result, we obtain the better performance rather than the conventional HOG. Especially the size of representation dimension is much smaller than the conventional HOG without reducing the detecting performance.
Scientific journal, English

Medical Image Processing and Computer-Aided Detection/Diagnosis (CAD)
H. Fujita; F. Nogata; H. Jiang; S. Kido; T. Feng; T. Hara; T. Hayashi; Y. Hirano; A. Katsumata; Y. Kawamura; T. Kokubo; J. Liu; C. Muramatsu; H. Shouno; R. Tachibana; X. Wang; F. Xiang; R. Xu; B. Yang; Y. Yokota; L. Zhang; Q. Li; Z. Guo
2012 INTERNATIONAL CONFERENCE ON COMPUTERIZED HEALTHCARE (ICCH), IEEE, 70-75, 2012, Peer-reviwed, Computer-aided detection/diagnosis (CAD) is emerging as an innovative interdisciplinary technology for medical service. The traditional concept of automated computer diagnosis is encountered with a significant barrier because computerized medical systems cannot fully replace human doctors with the comparable level of performance. By contrast, CAD is becoming widely adopted in clinical work because it offers complementary computing power to enhance doctor's competence for medical examination. 4 state-of- the-art CAD technologies were presented in the special session of medical image processing and CAD at ICCH 2012 as reported in this short paper. Those technologies will be briefly introduced here to show the current trend of development of CAD and to demonstrate how CAD helps in medical care.
International conference proceedings, English

An idiopathic interstitial pneumonia classification for CT image by use of a semi-supervised learning
Masayoshi Wada; Hayaru Shouno; Shoji Kido
Proceedings of International Forum on Medical Imaging in Asia (IFMIA) 2012, P1-34, 2012, Peer-reviwed
Symposium, English

Topographic Representations for Linearly Correlated Components
Hiroaki Sasaki; Michael Gutmann; Hayaru Shouno; Aapo Hyvarinen
Workshop on Deep Learning and Unsupervised Feature Learning, NIPS, 1, Dec. 2011, Peer-reviwed
International conference proceedings, English

Classification of idiopathic interstitial pneumonia CT images using convolutional-net with sparse feature extractors
Inagaki, T.; Shouno, H.; Kido, S.; Arabnia, H.R.
Proceedings of the 2011 International Conference on Parallel and Distributed Processing Techniques and Applications, 2, 699-705, Jul. 2011, Peer-reviwed
Scientific journal, English

Computer-Aided Diagnosis of Computational Anatomical Model and Application of Computer-Aided Autopsy Imaging
KIDO Shoji; HIRANO Yasushi; XU Rui; SHOUNO Hayaru
Medical Imaging Technology, The Japanese Society of Medical Imaging Technology, 29, 3, 138-142, 2011, Invited, The purposes of our research group in "computational anatomy for computer-aided diagnosis and therapy" are to develop and evaluate computer-aided diagnosis system based on computational anatomy, and to develop a method for approaching lifetime images which consist of living and autopsy images. So, we will research our theme in cooperation with other research groups. In this article, we will describe background and purpose of our research group, and will describe some of our current research results.
Scientific journal, Japanese

A Bayesian hyperparameter inference for Radon-transformed image reconstruction
Hayaru Shouno; Madomi Yamasaki; Masato Okada
International Journal of Biomedical Imaging, 2011, ID870252, 2011, Peer-reviwed, We develop a hyperparameter inference method for image reconstruction from Radon transform which often appears in the computed tomography, in the manner of Bayesian inference. Hyperparameters are often introduced in Bayesian inference to control the strength ratio between prior information and the fidelity to the observation. Since the quality of the reconstructed image is controlled by the estimation accuracy of these hyperparameters, we apply Bayesian inference into the filtered back-projection (FBP) reconstruction method with hyperparameters inference and demonstrate that the estimated hyperparameters can adapt to the noise level in the observation automatically. In the computer simulation, at first, we show that our algorithm works well in the model framework environment, that is, observation noise is an additive white Gaussian noise case. Then, we also show that our algorithm works well in the more realistic environment, that is, observation noise is Poissonian noise case. After that, we demonstrate an application for the real chest CT image reconstruction under the Gaussian and Poissonian observation noises. Copyright © 2011 Hayaru Shouno et al.
Scientific journal, English

Bayesian Image Restoration for Medical Images Using Radon Transform
Hayaru Shouno; Masato Okada
JOURNAL OF THE PHYSICAL SOCIETY OF JAPAN, PHYSICAL SOC JAPAN, 79, 7, 074004, Jul. 2010, Peer-reviwed, We propose an image reconstruction algorithm using Bayesian inference for Radon transformed observation data, which often appears in the field of medical image reconstruction known as computed tomography (CT). In order to apply our Bayesian reconstruction method, we introduced several hyper-parameters that control the ratio between prior information and the fidelity of the observation process. Since the quality of the reconstructed image is influenced by the estimation accuracy of these hyper-parameters, we propose an inference method for them based on the marginal likelihood maximization principle as well as the image reconstruction method. We are able to demonstrate a reconstruction result superior to that obtained using the conventional filtered back projection method.
Scientific journal, English

Classification of Idiopathic Interstitial Pneumonia on High-resolution CT Images using Counter Propagation Network
Y. Tanaka; H. Shouno; S. Kido
Proc. of PDPTA'10, 2, 652-657, Jul. 2010, Peer-reviwed
International conference proceedings, English

A Hyper-parameter Inference for Radon Transformed Image Reconstruction Using Bayesian Inference
Hayaru Shouno; Masato Okada
MACHINE LEARNING IN MEDICAL IMAGING, SPRINGER-VERLAG BERLIN, 6357, 26-+, 2010, Peer-reviwed, We propose an hyper-parameter inference method in the manner of Bayesian inference for image reconstruction from Radon transformed observation which often appears in the computed tomography. Hyper-parameters are often introduced in Bayesian inference to control the strength ratio between prior information and the fidelity to the observation. Since the quality of the reconstructed image is influenced by the estimation accuracy of these hyper-parameters, we apply Bayesian inference into the filtered back projection (FBP) reconstruction method with hyper-parameters inference, and demonstrate that estimated hyper-parameters can adapt to the noise level in the observation automatically.
International conference proceedings, English

Classification of patterns for diffuse lung diseases in thoracic CT images by AdaBoost algorithm
Masayuki Kuwahara; Shoji Kido; Hayaru Shouno
Progress in Biomedical Optics and Imaging - Proceedings of SPIE, 7260, 726037, 2009, Peer-reviwed, CT images are considered as effective for differential diagnosis of diffuse lung diseases. However, the diagnosis of diffuse lung diseases is a difficult problem for the radiologists, because they show a variety of patterns on CT images. So, our purpose is to construct a computer-aided diagnosis (CAD) system for classification of patterns for diffuse lung diseases in thoracic CT images, which gives both quantitative and objective information as a second opinion, to decrease the burdens of radiologists. In this article, we propose a CAD system based on the conventional pattern recognition framework, which consists of two sub-systems; one is feature extraction part and the other is classification part. In the feature extraction part, we adopted a Gabor filter, which can extract patterns such like local edges and segments from input textures, as a feature extraction of CT images. In the recognition part, we used a boosting method. Boosting is a kind of voting method by several classifiers to improve decision precision. We applied AdaBoost algorithm for boosting method. At first, we evaluated each boosting component classifier, and we confirmed they had not enough performances in classification of patterns for diffuse lung diseases. Next, we evaluated the performance of boosting method. As a result, by use of our system, we could improve the classification rate of patterns for diffuse lung diseases. © 2009 SPIE.
International conference proceedings, English

A Next Generation Modeling Environment PLATO: Platform for Collaborative Brain System Modeling
Shiro Usui; Keiichiro Inagaki; Takayuki Kannon; Yoshimi Kamiyama; Shunji Satoh; Nilton L. Kamiji; Yutaka Hirata; Akito Ishihara; Hayaru Shouno
NEURAL INFORMATION PROCESSING, PT 1, PROCEEDINGS, SPRINGER-VERLAG BERLIN, 5863, 84-+, 2009, To understand the details of brain function, a large scale system model that reflects anatomical and neurophysiological characteristics needs to be implemented. Though numerous computational models of different brain areas have been proposed, these integration for the development of a large scale model have not yet been accomplished because these models were described by different programming languages, and mostly because they used different data formats. This paper introduces a platform for a collaborative brain system modeling (PLATO) where one can construct computational models using several programming languages and connect them at the I/O level with a common data format. As an example, a whole visual system model including eye movement, eye optics, retinal network and visual cortex is being developed. Preliminary results demonstrate that the integrated model successfully simulates the signal processing flow at the different stages of visual system.
International conference proceedings, English

Extraction of Pulmonary Areas in Thoracic CT Images Including Pleural Effusion Using a Registration Method
角森昭教; 庄野逸; 木戸尚治
Japan Society of Medical Imaging Technology, 日本医用画像工学会, 26, 5, 338-346, Nov. 2008, Peer-reviwed, 胸水の貯留量を定量的かつ客観的に測定することは経過観察に有効な手段となるが、胸水を含むCT画像において、胸水領域は胸壁などの周辺組織とのCT値が類似しているため従来からの濃淡値に着目した手法での肺野領域の抽出は困難であった。そこで本研究では胸水の貯留量を測定するための前処理である肺野の抽出に対してテンプレートを用いた抽出手法を提案し抽出精度の評価を行った。本手法は、比較的高精度に抽出できる骨領域などを利用し、レジストレーションを行い、正常例より作成した肺野テンプレートを変形させることにより、肺野内の疾患に左右されにくい抽出手法となっている。本手法を10症例の胸水貯留を有する肺野領域において適用し、他の従来手法と比較した。そして、胸水貯留を含んだ肺野領域が抽出手法において、本提案手法が有効であることを示す。(著者抄録)
Scientific journal, Japanese

Comparison of two-dimensional with three-dimensional analyses for diffuse lung diseases from thoracic CT images
Y.Sugata; S.Kido; H.Shouno
Medical Imaging and Information Sciences, 医用画像情報学会, 25, 3, 43-47, Oct. 2008, Peer-reviwed, CT画像で撮影された瀰漫性肺疾患を含むデータ153例(疾患を含む症例130例、正常例23例)を対象に、特徴解析の手法として陰影パターンに着目し、二次元・三次元でのテクスチャ解析を行い複数の特徴量を算出し、それらの特徴量から組合わせ特徴ベクトルを作成し、瀰漫性肺疾患の識別を行った。二次元の特徴解析、三次元の特徴解析の比較を行った結果二次元・三次元共に濃淡ヒストグラムと差分統計量から算出した特徴量の組合わせ特徴ベクトルが選択された。識別率に関してはデータ全体の識別率において共に約90%で識別率には有意差がないことが示された。このため、瀰漫性陰影の特徴解析に関しては二次元の画像処理でも十分な精度が得られると考えられた。但し三次元のボリュームデータを扱うことにより肺全体を対象とした瀰漫性陰影の解析が可能で、病変の三次元的な分布の把握などには有用と思われた。
Scientific journal, Japanese

Construction of a Supporting System for Portable Terminal Devices using Middle-ware Server to Communicate with DICOM Service
Seto Yuichi; Kido Shoji; Shouno Hayaru
Journal of Life Support Engineering, The Society of Life Support Engineering, 20, 0, 122-122, 2008

Recent studies around the neocognitron
Hayaru Shouno
NEURAL INFORMATION PROCESSING, PART I, SPRINGER-VERLAG BERLIN, 4984, 1061-1070, 2008, Peer-reviwed, Neocognitron, which was proposed by Fukushima, is recently studied in several styles. In this paper, we introduce these studies from the both engineering and biological sides. From the engineering side, we discussed about the ability of the pattern classifier of the Neocognitron and relationship to the "convolutional net", which is recently well studied in the field of pattern recognition. From the biological side, we tried to explain the recent result of a biological experiment with the Neocognitron, and compare it with another model.
International conference proceedings, English

肋骨の変位とばねモデルを用いた胸部CT画像の経時的差分画像
S.Kido; H.Shouno
医用画像情報学会論文誌, 24, 4, 266-272, Dec. 2007, Peer-reviwed
Scientific journal, Japanese

ネオコグニトロンによる視覚腹側経路のモデル化
T.Yoshizuka; H.Shouno; H.Miyamoto; M.Okada; K.Fukushima
神経回路学会論文誌, Japanese Neural Network Society, 14, 4, 266-272, Dec. 2007, Peer-reviwed, It is known that the object recognition is processed in the ventral pathway of the visual system in humans and monkeys. The neocognitron that was proposed by Fukushima is a hierarchical neural network model for pattern recognition. In this paper, we show that the neocognitron can be regarded as a proper biological model of the ventral pathway. From the biological point of view, the model of the ventral pathway should satisfy the following conditions. The model should be hierarchical, the synaptic connections should spread locally, and each component in the hierarchy should be homogeneous. The network architecture of the neocognitron satisfies these conditions. Thus, we investigate the functional similarity between the neocognitron and the ventral pathway. We compared the response property of the neocognitron with that of IT cells. On comparing our results with those obtained by Logothetis et al., we found that the result were very similar qualitatively. Thus, we conclude that the neocognitron is a proper model for the ventral pathway.
Scientific journal, Japanese

Analysis of idiopathic interstitial pneumonia by self organization map on high-resolution computed tomography images
Goto, Y.; Shouno, H.; Kido, S.
Proceedings of the 2007 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA'07), 753-758, Jun. 2007, Peer-reviwed
Scientific journal, English

Temporal Subtracted Images of Thoracic CT by use of Displacements of Ribs and Elastic Object Model
KIDO Shoji; SHOUNO Hayaru
Medical Imaging and Information Sciences, MEDICAL IMAGING AND INFORMATION SCIENCES, 24, 4, 126-130, 2007, We have developed the temporal subtraction method of thoracic CT images by use of displacements of ribs and elastic object model. In the first step, we calculated the rotation parameters of the extracted rib centerlines those were extracted by a plane sweep method. In the next step, we estimated accuracy of registration for images by inputting displacements of the extracted ribs to elastic object model. In the third step, we registered one image to the other image with estimated parameters. Finally, subtracted images were obtained. We applied this method to 14 pairs of thoracic CT image series for evaluation of our method.
Japanese

Development of Computerized System for Selection of Similar Images from Different Patients for Image Subtraction of Chest Radiographs
ODA Nobuhiro; AOKI Takatoshi; OKAZAKI Hiroko; KAKEDA Shingo; KOUROGI Yukunori; YAHARA Katuya; SHOUNO Hayaru
Transactions of the Japanese Society for Medical and Biological Engineering : BME, Japanese Society for Medical and Biological Engineering, 44, 3, 435-444, 10 Sep. 2006, The purpose of this study was to develop a novel computerized scheme to automatically select similar chest radiographs for image subtraction of patients who have no previous chest radiographs and to assist the radiologist's interpretation by presenting "similar subtraction images". A large database of approximately 15,000 posteroanterior chest radiographs, which were diagnosed as normal, was used for searching similar images of different patients. First, in this scheme, two clinical parameters (age and sex) were used for selecting similar images. Next, 100 images of candidates in the database were selected according to similarity in height and the area of the lung field in the target image. We used quantitative measurement for searching similar images; namely, the correlation value of cheat region in the 100 images of the candidates. The similar subtraction images were obtained by subtracting the similar images selected from the target image. The performance of the proposed system was evaluated in comparison with 95 chest radiographs with a temporal subtraction image. The experimental results showed that the average of the correlation values in the temporal subtraction image and similar subtraction images were 0.9794 and 0.9574, respectively. Three radiologists subjectively evaluated various lung artifacts on the temporal subtraction image and similar subtraction images of 95 chest radiographs using a five-point rating scale (1:very poor, 2:poor, 3:adequate, 4:good, 5:excellent). Ratings higher than "adequate" were given for 70% of the similar subtraction images. This computerized scheme seems useful for the automatic selection of similar images for similar-image subtraction of chest radiographs and has potential use for assisting interpretations by radiologists.
Japanese

Naive mean field approximation for sourlas error correcting code
Masami Takata; Hayaru Shouno; Masato Okada
IEICE TRANSACTIONS ON INFORMATION AND SYSTEMS, IEICE-INST ELECTRONICS INFORMATION COMMUNICATIONS ENG, E89D, 8, 2439-2447, Aug. 2006, Peer-reviwed, Solving the error correcting code is an important goal with regard to communication theory. To reveal the error correcting code characteristics, several researchers have applied a statistical-mechanical approach to this problem. In our research, we have treated the error correcting code as a Bayes inference framework. Carrying out the inference in practice, we have applied the NMF (naive mean field) approximation to the MPM (maximizer of the posterior marginals) inference, which is a kind of Bayes inference. In the field of artificial neural networks, this approximation is used to reduce computational cost through the substitution of stochastic binary units with the deterministic continuous value units. However, few reports have quantitatively described the performance of this approximation. Therefore, we have analyzed the approximation performance from a theoretical viewpoint, and have compared our results with the computer simulation.
Scientific journal, English

Discrimination of Lung Sounds using a Statistics of Waveform Intervals
Taketoshi Orihashi; Hayaru Shouno; Shoji Kido
PDPTA'06, 2, 813-817, Jul. 2006
International conference proceedings, English

Classification of Idiopathic Interstitial Pneumonia using Boosting Method
Masayuki Kuwahara; Hayaru Shouno; Shoji Kido
Proc. of RSNA'05, 636, Dec. 2005, Peer-reviwed
International conference proceedings, English

声帯手術を目的としたコンピュータ支援診断インターフェイスの開発
井上陽介; 庄野逸; 木戸尚治; 後藤英功
医用画像情報学会誌, 22, 2, 160-167, Aug. 2005, Peer-reviwed
Scientific journal, Japanese

胸部単純X 線写真における他人による類似差分画像のための類似画像検索システムの開発
小田敍弘; 青木隆敏; 岡崎浩子; 掛田伸吾; 興絽征典; 矢原勝哉; 庄野逸
日本医用画像工学会誌, 23, 4, 250-258, Jul. 2005, Peer-reviwed
Scientific journal, Japanese

胸部単純X 線画像における結節性陰影抽出法の開発
日浦美香子; 木戸尚治; 庄野逸
日本医用画像工学会誌, The Japanese Society of Medical Imaging Technology, 22, 2, 160ー167-250, Jul. 2005, Peer-reviwed
Scientific journal, Japanese

Statistical Mechanics for Neural Spike Data Analysis by use of Log-Linear Model
Hayaru Shouno; Koji Wada; Masato Okada
Proc. of Randomness and Computation 2005, 58-59, Jul. 2005
International conference proceedings, English

Research on a classification of the medical data using the self-organization map
GOTO Yoshiharu; SHOUNO Hayaru; KIDO Shoji
IEICE technical report., The Institute of Electronics, Information and Communication Engineers, 104, 580, 157-161, 15 Jan. 2005, In the field of diagnosis using medical images, it is difficult to keep the quality of diagnosis with objectivity, since the diagnosis quality is affected by abilities of each doctor. Thus, the diagnosis aid system using computer is required for objective diagnosis in these decades. In this study, we try to construct a diagnosis aid system using a self-organizing map (SOM) algorithm, and classify several image data. The SOM is a kind of an unsupervised learning, in which high-dimensional input data embed into a 2-dimensional lattice structure called "map". In the embedding process, similarity between input data is used for assigning the data, and similar data is inclined to assign in near place in the map. Hence, we can easy to grasp the map, which describe the relationship of similarities among high-dimensional data, and we consider it as a important for making a objective diagnosis.
Japanese

Development of pulmonary nodule detection method on chest radiographs
Hiura, M.; Kido, S.; Shouno, H.
Medical Imaging Technology, 23, 4, 2005
Scientific journal

i-アプリを用いた数値計算の可能性
高田雅美; 柴山智子; 渡辺智恵美; 庄野逸; 城和貴
情報処理学会論文誌: 数理モデル化と応用, 46, 2, 47-55, Jan. 2005, Peer-reviwed
Scientific journal, Japanese

Statistical mechanics for neural spike data analysis using log-linear model
H Shouno; K Wada; M Okada
PROGRESS OF THEORETICAL PHYSICS SUPPLEMENT, PROGRESS THEORETICAL PHYSICS PUBLICATION OFFICE, 157, 157, 300-303, 2005, Peer-reviwed, Recently, we can simultaneously record spike data from many neurons in the field of electrophysiology, and thus it is required to develop mathematical framework for extracting higher-order correlation of neural firings. The joint probability of neural spike can be represented using the log-linear model. From statistical-mechanical point of view, the loglinear model can be regarded as a multi-body interacted Ising spin model or the Boltzman machine with higher-order interactions. The estimation of higher-order correlation of neural firing corresponds to that of higher-order interations in this Ising spin system, and to the hyper-parameter estimation in the Bayesian inference. In this paper, we apply maximization of marginal likelihood (MML) method to this problem, and discuss the properties of MML analytically using statistical-mechanical method.
Scientific journal, English

Development of Computer-Aided Diagnosis Scheme for Distinction between Benign and Malignant Pulmonary Nodules on Chest Radiographs Using Temporal Subtraction Images
ODA Nobuhiro; KIDO Shoji; SHOUNO Hayaru
Transactions of Japanese Society for Medical and Biological Engineering, Japanese Society for Medical and Biological Engineering, 42, 4, 209-214, 10 Dec. 2004, A novel automated computerized scheme has been developed to assist radiologists for distinction between benign and malignant pulmonary nodules on radiographs using temporal subtraction images. Fifty-one chest radiographs including 26 malignant nodules and 25 benign nodules were used. The CAD system was developed based on features extracted from both chest radiographs and temporal subtraction images. The nodule was segmented automatically on both chest radiographs and subtraction images once the location of the nodule was indicated on the chest radiograph by a radiologist and/or computer. The nodule on the subtraction image was then segmented by thresholding with various pixel values, which were determined from the area of the histogram of pixel values on the temporal subtraction image. Twenty-three image features for each nodule were obtained from both subtraction images and current chest radiographs. The nodule image features included three morphological features obtained from the subtraction image and 10 gray-level features obtained from a histogram analysis of pixel values within the nodule on both subtraction and current images. A linear discrimination analysis (LDA) with six features was applied to determine the likelihood of pulmonary nodule malignancy. A receiver operating characteristic (ROC) analysis was used in the area under the ROC curve (Az) of the computer output obtained by use of the LDA. The six image features selected were the area, irregularity, mean, squared mean, and contrast obtained from the subtraction image and contrast obtained from the current image, which provided the highest Az value of the computer output obtained using the LDA. LDA was employed to separate benign from malignant nodules by use of a hyperplane. The output value of LDA represented the distance of either a benign or a malignant nodule from the hyperplane. In fact, the Az value of the computer output with six features obtained using the LDA for distinction between benign and malignant nodules was 0.851, which was obtained from a leave-one-out method. Our CAD system has the potential to assist radiologists in distinguishing between benign and malignant pulmonary nodules on chest radiographs using temporal subtraction images.
Japanese

経時差分画像を用いた胸部単純写真における肺結節の良悪性鑑別のための自動化手法の開発
小田敍弘; 木戸尚治; 庄野逸
生体医工学, 42, 4, 9-14, Dec. 2004, Peer-reviwed
Scientific journal, Japanese

Accuracy of the Bethe approximation for hyperparameter estimation in probabilistic image processing
K Tanaka; H Shouno; M Okada; DM Titterington
JOURNAL OF PHYSICS A-MATHEMATICAL AND GENERAL, IOP PUBLISHING LTD, 37, 36, 8675-8695, Sep. 2004, Peer-reviwed, We investigate the accuracy of statistical-mechanical approximations for the estimation of hyperparameters from observable data in probabilistic image processing, which is based on Bayesian statistics and maximum likelihood estimation. Hyperparameters in statistical science correspond to interactions or external fields in the statistical-mechanics context. In this paper, hyperparameters in the probabilistic model are determined so as to maximize a marginal likelihood. A practical algorithm is described for grey-level image restoration based on a Gaussian graphical model and the Bethe approximation. The algorithm corresponds to loopy belief propagation in artificial intelligence. We examine the accuracy of hyperparameter estimation when we use the Bethe approximation. It is well known that a practical algorithm for probabilistic image processing can be prescribed analytically when a Gaussian graphical model is adopted as a prior probabilistic model in Bayes' formula. We are therefore able to compare, in a numerical study, results obtained through meanfield-type approximations with those based on exact calculation.
Scientific journal, English

Statistical Mechanics for Neural Spike Data Analysis using Log-Linear Model
Hayaru Shouno; Koji Wada; Masato Okada
Proc. of SPDSA'04, 79, Jul. 2004
International conference proceedings, English

Analysis of Bidirectional Associative Memory using Self-Consistent Signal to Noise Analysis and Statistical Neurodynamics
H. Shouno; S. Kido; M. Okada
Journal of Physics Society in Japan, THE PHYSICAL SOCIETY OF JAPAN, 73, 9, 2406-2412, Feb. 2004, Peer-reviwed, Bidirectional associative memory (BAM) is a kind of an artificial neural network used to memorize and retrieve heterogeneous pattern pairs. Many efforts have been made to improve BAM from the the viewpoint of computer application, and few theoretical studies have been done. We investigated the theoretical characteristics of BAM using a framework of statistical–mechanical analysis. To investigate the equilibrium state of BAM, we applied self-consistent signal to noise analysis (SCSNA) and obtained a macroscopic parameter equations and relative capacity. Moreover, to investigate not only the equilibrium state but also the retrieval process of reaching the equilibrium state, we applied statistical neurodynamics to the update rule of BAM and obtained evolution equations for the macroscopic parameters. These evolution equations are consistent with the results of SCSNA in the equilibrium state.
Scientific journal, English

Development of computerized system for detection of pulmonary nodules on digital chest radiographs using temporal subtraction images
Oda, N.; Kido, S.; Shouno, H.; Ueda, K.
Transactions of the Institute of Electronics, Information and Communication Engineers D-II, J87D-II, 1, 2004
Scientific journal

胸部単純X 線写真における経時差分画像を用いた結節状陰影の自動検出システムの開発
小田敍弘; 木戸尚治; 庄野逸; 上田克彦
電子情報通信学会論文誌, (一社)電子情報通信学会, 87, 1, 208-218, Jan. 2004, Peer-reviwed, 胸部単純X線写真の現在画像から過去画像を差分処理して得られる経時的差分画像によって肺結節状陰影を自動検出し,更にルールベース法やマハラノビス距離の判定法などを用いて偽陽性の数を減らす工夫を施したコンピュータ支援診断システムの開発を行った.比較的難易度の高いデータベースを用いた性能評価実験で,肺結節陰影の検出率80%,偽陽性数2.6(個/画像)と優れた性能を有することが確認された
Scientific journal, Japanese

Belief propagation for image restoration by using Gaussian model
Kazuyuki Tanaka; Noriko Yoshiike; Hayaru Shouno; Masato Okada
Proceedings of Sixth Workshop on Information-Based Induction Sciences, 83-88, Nov. 2003, Peer-reviwed
Research society, Japanese

擬似スペクトル法を用いた乱流場の直接数値シミュレーションの並列化と性能評価
高田雅美; 山本義暢; 庄野逸; 功刀資彰; 城和貴
情報処理学会論文誌: コンピューティングシステム, SIG6, 45-54, May 2003, Peer-reviwed
Scientific journal, Japanese

An Effective Evaluation Function for ICA to Separate Train Noise from Telluric Current Data
Mika Koganeyama; Sawa Sayuri; Hayaru Shouno; Toshiyasu Nagao; Kazuki Joe
Proc. of ICA'03, 837-842, May 2003, Peer-reviwed
International conference proceedings, English

Automatic segmentation of small pulmonary nodules on multidetector-row CT images
R Tachibana; S Kido; H Shouno; T Matsumoto
CARS 2003: COMPUTER ASSISTED RADIOLOGY AND SURGERY, PROCEEDINGS, ELSEVIER SCIENCE BV, 1256, 1389-1389, 2003, Peer-reviwed
International conference proceedings, English

Naive mean field approximation for image restoration
H Shouno; K Wada; M Okada
JOURNAL OF THE PHYSICAL SOCIETY OF JAPAN, PHYSICAL SOC JAPAN, 71, 10, 2406-2413, Oct. 2002, Peer-reviwed, We attempt image restoration in the framework of the Bayesian inference. Recently, it has been shown that under a certain criterion the MAP (Maximum A Posterior) estimate, which corresponds to the minimization of energy, can be outperformed by the MPM (Maximizer of the Posterior Marginals) estimate, which is equivalent to a finite-temperature decoding method. Since a lot of computational time is needed for the MPM estimate to calculate the thermal averages, the mean field method, which is a deterministic algorithm, is often utilized to avoid this difficulty. We present a statistical-mechanical analysis of naive mean field approximation in the framework of image restoration. We compare our theoretical results with those of computer simulation, and investigate the potential of naive mean field approximation.
Scientific journal, English

ICAを用いた地電流データからの電車ノイズと地震前兆シグナルの分離
小金山美賀; 庄野逸; 長尾年恭; 城和貴
情報処理学会論文誌: 数理モデル化と応用, Information Processing Society of Japan (IPSJ), 43, 7, 92-104, Sep. 2002, Peer-reviwed, Seismic electric signals (SESs) are sometimes contained in telluric current data (TCD). The method of detecting SESs in TCD has attracted notice recently as an effective method for short-term earthquake prediction. However, since most of the TCD collected in Japan is affected by train noise, therefore detecting SESs in TCD itself is considered as an extremely arduous job. The goal of our research is to obtain a method for detecting SESs, which is difficult because of train noises. The SES and train noise are considered as independent source signal. In this paper, we try to apply ICA (Independent Component Analysis) to several sets of TCDs and evaluate the results.
Scientific journal, Japanese

An Improvement of Program Partitioning Based Genetic Algorithm
Masami Takata; Hayaru Shouno; Kazuki Joe
Proc. of PDPTA'02, 1, 215-221, Jun. 2002, Peer-reviwed
International conference proceedings, English

Analysis of Bidirectional Associative Memory using SCSNA and Statistical Neurodynamics
Hayaru Shouno; Masato Okada
Proc. of PDPTA'02, 1, 239-245, Jun. 2002, Peer-reviwed
International conference proceedings, English

The Design and Implementation of Unimodular Transformations for the Parallelizing Compiler PROMIS
Hisako Ishiuchi; Tomomi Yamaguchi; Hayaru Shouno; Kazuki Joe
Proc. of PDPTA'02, 1, 1438-1443, Jun. 2002, Peer-reviwed
International conference proceedings, English

Classification of Visualized Data Dependence
Asami Iwasaka; Hayaru Shouno; Mariko Sasakura; Kazuki Joe
Proc. of PDPTA'02, 1, 1444-1450, Jun. 2002, Peer-reviwed
International conference proceedings, English

Parallelization of Seismic Wave Calculation by Impulse Response Functions
Kyoko Fukuda; Toshihiko Hayasaka; Hayaru Shouno; Kazuki Joe
Proc. of PDPTA'02, 1, 1465-1474, Jun. 2002, Peer-reviwed
International conference proceedings, English

Separation of Train Noise and Seismic Electric Signals from Telluric Current Data by ICA
Mika Koganeyama; Hayaru Shouno; Toshiyasu Nagao; Kazuki Joe
Proc. of ICA'01, 367-372, Dec. 2001, Peer-reviwed
International conference proceedings, English

Formation of a direction map by projection learning using Kohonen's self-organization map
H Shouno; K Kurata
BIOLOGICAL CYBERNETICS, SPRINGER-VERLAG, 85, 4, 241-246, Oct. 2001, Peer-reviwed, In this paper, we propose a modification of Kohonen's self-organization map (SOM) algorithm. When the input signal space is not convex, some reference vectors of SOM can protrude from it. The input signal space must be convex to keep all the reference vectors fixed on it for any updates. Thus, we introduce a projection learning method that fixes the reference vectors onto the input signal space. This version of SOM can be applied to a non-convex input signal space. We applied SOM with projection learning to a direction map observed in the primary visual cortex of area 17 of ferrets, and area 18 of cats. Neurons in those areas responded selectively to the orientation of edges or line segments, and their directions of motion. Some iso-orientation domains were subdivided into selective regions for the opposite direction of motion. The abstract input signal space of the direction map described in the manner proposed by Obermayer and Blasdel [ (1993) J Neurosci 13: 4114-4129] is not convex. We successfully used SOM with projection learning to reproduce a direction-orientation joint map.
Scientific journal, English

An Improvement of Program Partitioning Based Genetic Algorithm
Masami Takata; Hayaru Shouno; Kazuki Joe
Proc. of HPC Asia'01, CD-ROM 16pages, Sep. 2001, Peer-reviwed
International conference proceedings, English

Task and variable representation graph: an intermediate representation of parallelizing compilers for distributed shared memory systems
Haneda, M.; Shouno, H.; Joe, K.
JSPP2001. Joint Symposium on Parallel Processing 2001, 2001
Scientific journal

Statistical-mechanical approach for analog neural network model used in image restoration
H Shouno; M Okada
PDPTA'2001: PROCEEDINGS OF THE INTERNATIONAL CONFERENCE ON PARALLEL AND DISTRIBUTED PROCESSING TECHNIQUES AND APPLICATIONS, C S R E A PRESS, 3, 1298-1304, 2001, Peer-reviwed, The ability to restore an image front signals received through a noisy channel is an important concern. This issue is related to the physics theory of spin-glass. In the theory, the Ising spin system is usually used for image restoration, however, a lot of calculation time is needed to obtain precise solution. As a result many researchers substitute the Ising spin model with the analog neural network model. We analyzed the analog neural network ability applied to the image restoration problem using the mean field theory. With the conventional image restoration method, the estimated overlap with the analog neural network model is equivalent to that of the Ising spin model. When parity codes are sent, the analog neural network's ability does not improve over the Ising spin model. If the noise variable is small, however, the performance of the analog neural network model is as good as the Ising spin model.
International conference proceedings, English

Detecting seismic electric signals by LVQ based clustering
K Fukuda; M Koganeyama; H Shouno; K Joe; T Nagao
PDPTA'2001: PROCEEDINGS OF THE INTERNATIONAL CONFERENCE ON PARALLEL AND DISTRIBUTED PROCESSING TECHNIQUES AND APPLICATIONS, C S R E A PRESS, 3, 1305-1311, 2001, Peer-reviwed, Aiming at short-term prediction of earthquakes, we have proposed the use of neural networks for analyzing telluric current data observed by the VAN method. We have already tried a telluric CUM-Cat data analysis method with learning Vector Quantization. In this paper, we will show preliminary experimental results for categorization of telluric current data by its frequency for the Izu islands earthquakes in Japan.
International conference proceedings, English

The design and implementation of a UIR interface for the MIRAI parallelizing compiler
T Yamaguchi; H Shouno; K Joe
PDPTA'2001: PROCEEDINGS OF THE INTERNATIONAL CONFERENCE ON PARALLEL AND DISTRIBUTED PROCESSING TECHNIQUES AND APPLICATIONS, C S R E A PRESS, 1200-1206, 2001, Peer-reviwed, Since UIRs (Universal Intermediate Representations) for parallelizing compilers are a complicated arid the data set involved is quite large. Implementation, porting and maintenance of various optimizations are quite difficult, To overcome this problem, we propose a set of standard operations to a UIR of a parallelizing compiler as a UIR interface. In this paper, the specification and implementation issues of the UIR interface arc presented, and implementation examples of several tool, transformation methods are given and evaluated.
International conference proceedings, English

Handwritten digit recognition by a neocognitron with improved bend-extractors
Fukushima, K.; Kimura, E.; Shouno, H.; Heiss, M.
Proceedings of NC 1998. International ICSC/IFAC Symposium on Neural Computation, 240-246, Sep. 1998, Peer-reviwed
Scientific journal, English

Neocognitron with improved bend-extractors
K Fukushima; E Kimura; H Shouno
IEEE WORLD CONGRESS ON COMPUTATIONAL INTELLIGENCE, IEEE, 1172-1175, 1998, Peer-reviwed, We have reported previously that the performance of a neocognitron can be improved by a built-in bend-extracting layer The conventional bend-extracting layer can detect bend points and end points of lines correctly, but not always crossing points of lines. This paper discusses that an introduction of a mechanism of disinhibition can make the bend-extracting layer detect not only bend points and end points but also crossing points of lines correctly. A neocognitron with this improved bend-extracting layer can recognize handwritten digits in the real world with a recognition rate of 98%.
International conference proceedings, English

Neocognitron with improved bend-extractors: Recognition of handwritten digits in the real world
K Fukushima; E Kimura; H Shouno
NEURAL COMPUTING & APPLICATIONS, SPRINGER, 7, 3, 260-272, 1998, Peer-reviwed, We have reported previously that the performance of a neocognitron can be improved by a builtin bend-extracting layer. The conventional bend-extracting layer can detect bend points and end points of lines correctly, but not always crossing points of lines. This paper shows that an introduction of a mechanism of disinhibition can make the bend-extracting layer detect not only bend points and end points, but also crossing points of lines correctly. This paper also demonstrates that a neocognitron with this improved bend-extracting layer can recognise handwritten digits in the real world with a recognition rate of about 98%. We use the technique of dual thresholds for feature-extracting S-cells, and higher threshold values are used in the learning than in the recognition phase. We discuss how the threshold values affect the recognition rate.
Scientific journal, English

射影学習を用いたＫｏｈｏｎｅｎ ｍｏｄｅｌによるＤｉｒｅｃｔｉｏｎ Ｍａｐの形成
Hayaru Shouno; Koji Kurata
Nihon Shinkei Kairo Gakkaishi, Japanese Neural Network Society, 4, 3, 109-114, May 1997, Peer-reviwed
Scientific journal, Japanese

Training neocognitron to recognize handwritten digits in the real world
K Fukushima; K Nagahara; H Shouno
SECOND AIZU INTERNATIONAL SYMPOSIUM ON PARALLEL ALGORITHMS/ARCHITECTURE SYNTHESIS, PROCEEDINGS, I E E E, COMPUTER SOC PRESS, 292-298, 1997, Using a large scale real-world database ETL-1, we show that the neocognitron trained by unsupervised learning with a winner-take-all process can. recognise handwritten digits with a recognition rate higher than 97%. We use the technique of dual thresholds for feature extracting S-cells, and higher threshold values are used in the learning than in the recognition phase. We discuss how the threshold values affect the recognition, rate. The learning method for the cells of the highest stage of the network has been, modified from the conventional one, in order to reconcile the unsupervised learning with the use of information of the category names of the training patterns.
International conference proceedings, English

Training Neocognitron to Recognize Handwritten Digits in the Real World
Kunihiko Fukushima; Kenichi Nagahara; Hayaru Shouno; Masato Okada
Proc. of WCNN'96, INNS Press, 1, 405-409, Sep. 1996
International conference proceedings, English

Handwritten digit recognition with a neocognitron using different thresholds in learning and recognition
Shouno, H.; Nagahara, K.; Fukushima, K.; Okada, M.; Amari, S.-I.; Xu, L.; Chan, L.-W.; King, I.; Leung, K.-S.
Progress in Neural Information Processing. Proceedings of the International Conference on Neural Information Processing, 1, 292-298, Sep. 1996
Scientific journal, English

Connected character recognition in cursive handwriting using selective attention model with bend processing
Hayaru Shuono; Kunihiko Fukushima
Syst. Comp. Jpn., Wiley-Blackwell, 26, 10, 35-46, 1995
Scientific journal

折れ点処理回路を用いた選択的注意機構による英字筆記体連結文字列認識
庄野逸; 福島邦彦
電子情報通信学会論文誌, The Institute of Electronics, Information and Communication Engineers, J77-DII, 5, 940-950, May 1994, Peer-reviwed, 手書き英字筆記体連結文字列認識を行うシステムの一つとして選択的注意機構を用いたシステムが今川らによって提唱された.しかしながら今川らのシステムは,それほど高い認識能力をもっていたわけではなく,認識する文字カテゴリーも5文字と比較的小規模なシステムであった.本研究では今川らの認識システムを拡張し,更に高い認識能力をもつシステムを作成した.我々は"選択的注意のモデル"の一部分がパターン認識システム"ネオコグニトロン"に類似していることに着目した.ネオコグニトロンにおいて,折れ点検出回路を導入すると認識能力の向上が認められることが報告されているので,我々は今川らのシステムに折れ点処理回路を導入したシステムを作成した.更に本システムに対して種々のテストパターンを与え,コンピュータシミュレーションを行い,筆記体連結文字列の認識に対して有効であることを確認した.
Scientific journal, Japanese

Cursive word recognition using selective attention with bend-processing
Fukushima, K.; Shouno, H.; Marinaro, M.; Morasso, P.G.
ICANN '94. Proceedings of the International Conference on Artificial Neural Networks, 2, 957-961, May 1994, Peer-reviwed
Scientific journal, English

Connected character recognition in cursive handwriting using attention model with bend processing
Shouno, H.; Fukushima, K.
Transactions of the Institute of Electronics, Information and Communication Engineers D-II, J77D-II, 5, 1994
Scientific journal

VISUAL-PATTERN RECOGNITION WITH SELECTIVE ATTENTION
K FUKUSHIMA; H SHOUNO
WORLD CONGRESS ON NEURAL NETWORKS-SAN DIEGO - 1994 INTERNATIONAL NEURAL NETWORK SOCIETY ANNUAL MEETING, VOL 1, LAWRENCE ERLBAUM ASSOC PUBL, 1, A575-A580, 1994, Peer-reviwed
International conference proceedings, English

Revealing the Mechanism of Large-scale Gradient Systems Using a Neural Reduced Potential
Shunya Tsuji; Ryo Murakami; Hayaru Shouno; Yoh-ichi Mototake
Dec. 2023, NeurIPS 2023 Workshop: Machine Learning and the Physical Sciences, Peer-reviwed, Summary international conference

Editorial for “A Lightweight Convolutional Neural Network Based on Dynamic Level‐Set Loss Function for Spine MR Image Segmentation”
Hayaru Shouno; Tomohisa Okada
Wiley, 27 Jun. 2023, Journal of Magnetic Resonance Imaging, 59, 4, 1454-1455, Invited, Introduction scientific journal, 1053-1807, 1522-2586, 137622878

スパースモデリングを用いた析出物画像からのメタル温度推定
遠藤瑛泰; 澤田浩太; 永田賢二; 吉川英樹; 庄野逸
日本神経回路学会, 05 Mar. 2021, 神経回路学会論文誌, 29, 1, 15-23, Japanese, Invited

A historical view of AI and its prospect
Hayaru Shouno
Medical View, 26 Oct. 2019, Clinical Imagiology, 35, 10, 1112-1119, Japanese, Invited, Introduction scientific journal

Application of Machine Learning to the Medical Image Processing
Hayaru Shouno
画像電子学会, 2018, 画像電子学会誌, 47, 4, 479-484, Japanese, Invited

脳情報科学が拓くAIとICT：2．脳情報科学と人工知能 -ネオコグニトロンからDeep Learningへ-
本武陽一; 庄野逸; 田村弘; 岡田真人
日本情報処理学会, 01 Dec. 2017, 情報処理, 59, 1, 42-47, Japanese, Invited, Introduction scientific journal

ディープラーニングの基礎とその関連技術
庄野 逸
In this paper, we explain about a basic architecture and learning style of deep convolution neural network (DCNN), which is known as a kind of deep learning (DL) system, and also show an application of medical image classification. The DCNN is a combination of neural network architecture called “Neocognitron” and learning method called error back propagation (BP). One of the important factor for the performance of DCNN is a balance between the number of the free parameters in the network and the scale of the training dataset. In several field such like medical imaging, it is hard to acquire labeled data. The small dataset sometimes occur the overtraining. In order to prevent the overtraining, we introduce a transfer style learning method into the DCNN, which improves the classification performance., The Japanese Society of Medical Imaging Technology, 30 Oct. 2017, Medical Imaging Technology, 35, 4, 180-186, Japanese, Invited, 130006108052

ディープラーニングの概要と医療分野への応用
庄野 逸
25 Sep. 2017, 月刊インナービジョン, 32, 7, 7-9, Japanese, Invited, Introduction commerce magazine

2 段階転移学習を用いたディープコンボリューションネットの医用画像認識
Hayaru Shouno; Aiga Suzuki; Satoshi Suzuki; Shoji Kido
本稿では，コンピュータビジョンの業界において既にデファクトスタンダードとなりつつあるディープコンボリューションネット（Deep Convolutional Neural Network：DCNN）の基本的な構造と学習様式を解説し，医療応用への一例を述べる．DCNNは，ネットワークアーキテクチャをFukushimaのネオコグニトロン（Fukushima, K., Biological Cybernetics, Vol.36, No.4, pp.193-202, 1980）として，学習手法を誤差逆伝搬（Error Back Propagation：BP）法を適用した手法であり，比較的な古典的なアーキテクチャと学習手法で構築されている．これらの手法は1980年代から存在するが，もっとも変革が大きい部分は，学習データセットの質と量の変化である．DCNNに代表されるディープラーニングにおいて重要なのは，システムの内部表現の自由度と学習サンプルとの兼ね合いであり，ビッグデータ時代に突入した現在において単純な写真等のデータを確保するのは比較的楽にできるようになってきている．その一方で，医療分野などの計測にコストが掛かるような領域では，学習サンプルを如何に確保するかは重要な問題になると考えられる．我々は，このような少数データセットへの学習方式として転移型の学習様式を用いて，DCNNを構築することを提案し，一定の成果を上げることに成功した．, Japanese Neural Network Society, May 2017, 神経回路学会論文誌, 24, 1, 3-12, Japanese, 1340-766X, 130006832269

A 2-staged Transfer Learning Method with Deep Convolutional Neural Network for Diffuse Lung Disease Analysis
Aiga Suzuki; Satoshi Suzuki; Hayaru Shouno; Shoji Kido
Jan. 2017, Proc. of IFMIA 2017, -, -, ---, English

ディープラーニングの医用画像への応用
Hayaru Shouno; Satoshi Suzuki; Shoji Kido
医用画像情報学会, Dec. 2016, 医用画像情報学会誌, 33, 4, 75-80, Japanese, Invited, 0910-1543, 1880-4977, 2017361460

Basic Technology for Sparse Modeling with Some Historical View
Hayaru Shouno
Institute of Electronics, Information and Communication Enginners, Jul. 2016, The Journal of Institute of Electronics, Information and Communication Enginners, 99, 5, 376-380, Japanese, Invited, 2188-2355

Basic Technology for Sparse Modeling with Some Historical View
庄野 逸
電子情報通信学会, May 2016, 電子情報通信学会誌 = The journal of the Institute of Electronics, Information and Communication Engineers, 99, 5, 376-380, Japanese, 0913-5693, 40020842488, AN1001339X

22aBT-7 Study on role of dropout as a regularizer
Kondou T.; Suzuki S.; Saitoh D.; Hara K.; Shouno H.
The Physical Society of Japan (JPS), 2016, Meeting Abstracts of the Physical Society of Japan, 71, 3095-3095, Japanese, 2189-079X, 110010059010, AA12721570

22aBL-1 Study on convergence property of on-line learning using convolution network
Saito Daisuke; Hara Kazuyuki; Shouno Hayaru
The Physical Society of Japan (JPS), 2015, Meeting Abstracts of the Physical Society of Japan, 70, 2989-2989, Japanese, 2189-079X, 110009992263, AA12721570

スパースモデリングと topograhic ICA
庄野 逸; 佐々木 博昭
映像情報メディア学会, Dec. 2014, 映像情報メディア学会誌, 68, 12, 888-891, Japanese, Invited, 1342-6907, 110009892560, AN10588970

特集/統計的画像処理の技術動向-序文
Hayaru Shouno
日本医用画像工学会, May 2014, Medical Imaging Technology, 32, 4, 153-154, Japanese, Invited

局所画像特徴量〜SIFT, HOGを題材に〜
庄野 逸
映像情報メディア学会, Mar. 2013, 映像情報メディア学会誌, 67, 3, 256-258, Japanese, Invited, 1342-6907, 110009597687, AN10588970

Platform for collaborative brain system modeling (PLATO)
K. Inagaki; T. Kannon; Y. Kamiyama; S. Sato; N. Kamiji; Y. Hirata; A. Ishihara; H. Shouno; S. Usui
Jun. 2009, Society for Neuroscience, -, -, ---, English, Peer-reviwed

Platform for collaborative brain system modeling (PLATO): toward large scale modeling for visual system
Shiro Usui; Takayuki Kannon; Yoshimi Kamiyama; Keiichiro Inagaki; Shunji Satoh; Yutaka Hirata; Nilton Kamiji; Akito Ishihara; Hayaru Shouno
Frontiers Media {SA}, 1970, Front. 2nd. INCF. Congr. of. Neuro., 15247412

2020-2021年版 医用画像ディープラーニング入門
庄野 逸
Scholarly book, Japanese, Joint work, 第1章 人工知能総論, オーム社, 14 May 2020

AI 辞典 第3版
庄野 逸
Scholarly book, Japanese, Single work, 6.10 節 エッジコンピューティングと人工知能 13.5節 深層学習と視覚, 近代科学社, 21 Dec. 2019

Deep Learning in Healthcare
A. Suzuki; H. Sakanashi; S. Kido; H. Shouno
Scholarly book, English, Joint work, Chpater: Deep Learning in Textural Medical Image Analysis, Springer, Cham, 19 Nov. 2019

人と共生するAI革命ー活用じれいからみる生活・産業・社会の未来展望ー
Hayaru Shouno
Scholarly book, Japanese, Joint work, 第3章3節, 株式会社エヌ・ティー・エス, 10 Jun. 2019, 9784860436087

医用画像ディープラーニング入門
Scholarly book, Japanese, Joint work, 第1章 人工知能総論, オーム社, 10 Apr. 2019, 9784274223655

AI白書2017～人工知能がもたらす技術の革新と社会の変貌～
独立行政法人情報処理推進機構; AI白書編集委員会
General book, Japanese, Joint work, 1.2 .3節 脳における視覚系のモデル, 角川アスキー総合研究所, 22 Jul. 2017, 9784048996075

実践医用画像解析ハンドブック
Dictionary or encycropedia, Japanese, Joint work, オーム社, 01 Nov. 2012, 9784274212826

Knowledge Based Intelligent Techniques
H. Shouno; K. Fukushima; M. Okada
English, Joint work, Chapter2: Recognition of Handwritten Digits in the Real World by Neocognitron, CRC Press, 1998

線形予測に基づくロスレス音声コーデックSRLAの圧縮率改善
峰尾 太陽; 庄野 逸
Oral presentation, Japanese, 情報処理学会 音楽情報科学研究会
27 Aug. 2024
26 Aug. 2024- 27 Aug. 2024

X線光電子分光スペクトルにおける共通ピーク構造のベイズ推定
丸山颯斗; 村上諒; 篠塚寛志; 永田賢二; 吉川英樹; 庄野逸
Oral presentation, Japanese, 情報処理学会数理モデル化と問題解決研究会
22 Jun. 2024
20 Jun. 2024- 22 Jun. 2024

Precipitate segmentation for metal temperature estimation using deep learning
Yuki Sasaki; Akihiro Endo; Kota Sawada; Kenji Nagata; Hayaru Shouno
Oral presentation, English, MRM2023
14 Dec. 2023
11 Dec. 2023- 16 Dec. 2023

System Development for Correlation Analysis with Measurement Meta-data and Spectral Structures in XPS data
Ryo Murakami; Kenji Nagata; Hayaru Shouno; Hiroshi Shinotsuka; Hideki Yoshikawa
Oral presentation, English, MRM2023
14 Dec. 2023
10 Dec. 2023- 16 Dec. 2023

Compaction of a Transformer Model for Finding Correspondences in Images
Hideo Terada; Hayaru Shouno
Oral presentation, Japanese, Compaction of a Transformer Model for Finding Correspondences in Images
30 Jun. 2023
29 Jun. 2023- 01 Jul. 2023

Proposal for an analytical framework for gradient systems using Neural reduced potential
Shunya Tsuji; Ryo Murakami; Hayaru Shouno; Yo-ichi Mototake
Oral presentation, Japanese, 第37回日本人工知能学会全国大会
07 Jun. 2023
06 Jun. 2023- 09 Jun. 2023

Grad-CAM approach for Multiclass Magnetic Resonance Imaging Tumor detection and Classification
Tahir Hussain; Hayaru Shouno
Oral presentation, English, 電子情報通信学会メディカルイメージング研究会
18 May 2023

ロスレス音声符号化のためのSVRによるFIRシステム同定
峰尾 太陽; 庄野 逸
Oral presentation, Japanese, 情報処理学会数理モデル化と問題解決研究会
09 Mar. 2023
09 Mar. 2023- 10 Mar. 2023

VGGモデルの視覚野的解釈における解析
樋口 陽光; 寺元 陶冶; 鈴木 聡志; 庄野 逸
Oral presentation, Japanese, 情報処理学会数理モデル化と問題解決研究会
09 Mar. 2023
09 Mar. 2023- 09 Mar. 2023

少数データを用いたAnti-Aliased Convolutional Neural Network構築のための知識蒸留学習
鈴木 聡志; 武田 翔一郎; 澤田 雅人; 増村 亮; 庄野 逸
Poster presentation, Japanese, Neuro2022, 日本神経科学会，日本神経化学会，日本神経回路学会, 沖縄コンベンションセンター, Domestic conference
02 Jul. 2022

Portilla-Simoncelli statistics を用いたResNetのテクスチャ解析
長坂 祥子; 庄野 逸
Oral presentation, Japanese, Neuro2022, 日本神経科学会，日本神経化学会，日本神経回路学会, 沖縄コンベンションセンター, Domestic conference
01 Jul. 2022

圧縮センシングを用いたXMCD-CT再構成
滝澤月斗; 庄野 逸; 水牧仁一朗; 鈴木基寛
Oral presentation, Japanese, 電子情報通信学会ニューロコンピューティング研究会, 電子情報通信学会, オンライン, Domestic conference
03 Mar. 2022

Adversarial Training with Knowledge Distillation considering INtermediate Feature Representation in CNNs
Hikaru Higuchi; Satoshi Suzuki; Hayaru Shouno
Oral presentation, Japanese, 電子情報通信学会ニューロコンピューティング研究会, 電子情報通信学会, オンライン, Domestic conference
14 Jan. 2022

A lossless audio codec based on hierarchical residual prediction
Taiyo Mineo; Hayaru Shouno
Oral presentation, Japanese, 電子情報通信学会SIP研究会, 電子情報通信学会, オンライン, 本研究では，Neural Network (NN) の高い予測精度を保ちつつもデコード負荷を低く抑えたモデルを使用したロスレス音声コーデックを提案する．提案モデルは，音声を一定長のフレームで切り出し，その範囲で残差の符号長が短くなるように係数を補助関数法によって設定する．このモデルは残差を繰り返し予測するため，ResNetと同様の構成を持つ．提案手法をコーデックとして実装し，性能比較実験を行った．圧縮率においてMonkey's Audioを除き高い圧縮率を示し，デコード速度は実用的であることを示した．, Domestic conference
13 Jan. 2022

視覚認知に基づく画像統計量を⽤いた磁区パターン解析
辻 駿哉; 村上 諒; 水牧 仁一郎; 赤井 一郎; 庄野 逸
Oral presentation, Japanese, 第35回日本放射光学会年会 放射光科学合同シンポジウム, 日本放射光学会, オンライン, http://www.jssrr.jp/jsr2022/, Domestic conference
08 Jan. 2022

Magnetic Domain Texture Pattern Analysis Using Wavelet-Based Joint Texture Statistics
Ryo Murakami; Masaichiro Mizumaki; Ichiro Akai; Hayaru Shouno
Oral presentation, English, Materials Research Meeting 21, 日本MRS, パシフィコ横浜, https://confit.atlas.jp/guide/event/mrm2021/top, International conference
13 Dec. 2021

Automatic Estimation of XPS Reference Spectra for TiO2 Semiconductor Free from Equipment-derived Arbitrariness
Ryo Murakami; Kenji Nagata; Hideki Yoshikawa; Hiroshi Shinotsuka; Hayaru Shouno
Oral presentation, English, Materials Research Meeting 21, 日本MRS, パシフィコ横浜, https://confit.atlas.jp/guide/event/mrm2021/top, International conference
13 Dec. 2021

Prediction of metal temperature in creep-exposed austenitic stainless steel from optical micrographs with sparse regression method
Akihiro Endo; Kota Sawada; Kenji Nagata; Hideki Yoshikawa; Hayaru Shouno
Oral presentation, English, Materials Research Meeting 21, 日本MRS, パシフィコ横浜, https://confit.atlas.jp/guide/event/mrm2021/top, International conference
13 Dec. 2021

機械学習を用いた光学顕微鏡写真からのメタル温度予測
遠藤 瑛泰; 澤田 浩太; 永田 賢二; 吉川 英樹; 庄野 逸
Oral presentation, Japanese, 日本金属学会第169回講演大会, 電子金属学会, Domestic conference
16 Sep. 2021

受容野の最適刺激を用いた畳込みニューラルネットワークの可視化手法
Genta Kobayashi; Hayaru Shouno
Oral presentation, Japanese, 電子情報通信学会 ニューロコンピューティング研究会, 電子情報通信学会, Domestic conference
03 Mar. 2021

Portilla-Simoncelli Statisticsを用いたDCNNのテクスチャ特徴解析
Yusuke Hamano; Hayaru Shouno
Oral presentation, Japanese, 電子情報通信学会 ニューロコンピューティング研究会, 電子情報通信学会, Domestic conference
03 Mar. 2021

視覚野構造に基づいたシフト不変な深層学習モデルの確立
樋口 陽光; 鈴木 聡志; 庄野 逸
Oral presentation, Japanese, 第30回日本神経回路学会全国大会, 日本神経回路学会, バーチャル, Domestic conference
02 Dec. 2020

BIC自動ピークフィッティング技術を用いたXPSデータセットのハイスループット解析手法
永田 賢二; 角谷 正友; 篠塚 寛志; 田沼 繁夫; 登坂 弘明; 原田 善之; 松波 成行; 吉川 英樹; 庄野逸; 村上諒
Oral presentation, Japanese, 2020年度 実用表面分析講演会, 表面分析研究会, Domestic conference
24 Nov. 2020

情報量規準を用いた信頼区間推定付きのXPSスペクトルの自動解析
篠塚 寛志; 永田 賢二; 吉川 英樹; 本武 陽一; 庄野 逸; 岡田 真人
Oral presentation, Japanese, 2020年日本表面真空学会学術講演会, 日本表面真空学会, https://doi.org/10.14886/jvss.2020.0_164, Domestic conference
19 Nov. 2020

参照スペクトルを使った多元素XPSスペクトルの解析手法の開発
村上 諒; 庄野 逸; 永田 賢二; 篠塚 寛志; 吉川 英樹
Oral presentation, Japanese, 2020年日本表面真空学会学術講演会, 日本表面真空学会, https://doi.org/10.14886/jvss.2020.0_165, Domestic conference
19 Nov. 2020

多量のスペクトルデータを利用した参照スペクトルの推定手法の開発
村上 諒; 庄野 逸; 篠塚 寛志; 永田 賢二; 吉川 英樹
Oral presentation, Japanese, 第81回日本応用物理学会秋季学術講演会, 日本応用物理学会, Domestic conference
09 Sep. 2020

多量のスペクトルデータを利用した参照スペクトルの推定手法の開発
篠塚 寛志; 永田 賢二; 吉川 英樹; 本武 陽一; 庄野 逸; 岡田 真人
Oral presentation, Japanese, 第81回日本応用物理学会秋季学術講演会, 日本応用物理学会, Domestic conference
09 Sep. 2020

Improvement Convergence Rate of the Sign Algorithm by Natural Gradient Method
Taiyo Mineo; Hayaru Shouno
Oral presentation, Japanese, 電子情報通信学会 信号処理研究会, 電子情報通信学会, Domestic conference
Aug. 2020

スパースコーディングを用いた惑星表面画像のための圧縮手法の提案
Yoshifumi Uesaka; Hayaru Shouno
Oral presentation, Japanese, 電子情報通信学会 ニューロコンピューティング研究会, 電子情報通信学会, 宮古島マリンターミナル, Domestic conference
24 Jan. 2020

スパース推定を用いた潜在的な犯罪の高リスクエリアの推定と犯罪発生メカニズムの考察
Sho Ichigozaki; Takahiro Kawashima; Hayaru Shouno
Oral presentation, Japanese, 電子情報通信学会 ニューロコンピューティング研究会, 電子情報通信学会, 宮古島マリンターミナル, Domestic conference
24 Jan. 2020

Implementation of an FPGA-based energy-efficient MCMC method for 2D Lenz-Ising model
Patrick Tchicali; Hayaru Shouno
Oral presentation, English, 電子情報通信学会 ニューロコンピューティング研究会, 電子情報通信学会, 豊橋技術科学大学, Domestic conference
06 Dec. 2019

Bolasso特徴選択手法を用いたびまん性肺疾患陰影の分析
Akihiro Endo; Kenji Nagata; Shoji Kido; Hayaru Shouno
Oral presentation, Japanese, 電子情報通信学会 ニューロコンピューティング研究会, 電子情報通信学会, 東北大学, Domestic conference
04 Oct. 2019

TV正則化と辞書学習を用いたOS-EM法におけるPET画像再構成
Naohiro Okumura; Hayaru Shouno
Poster presentation, Japanese, The 38th JAMIT Annual Meeting, 日本医用画像工学会, 奈良春日野国際フォーラム, http://jamit2019.jamit.jp/, Domestic conference
26 Jul. 2019

Bolasso を用いたびまん性肺疾患画像の特徴選択
遠藤 瑛泰; 永田 賢二; 木戸 尚治; 庄野 逸
Oral presentation, Japanese, 情報処理学会数理モデルかと問題解決研究会, 情報処理学会, 沖縄科学技術大学院大学, Diffuse lung disease is diseases with abnormal shadows on lung CT images, and requires early detection and appropriate treatment. These shadows indicate the nature of the lesion and provide clues to the diagnosis such as identification of the disease and confirmation of the progression. Therefore, we tried to selrct Features which express shadows well from features extracted from images and interpret shadows. In this paper, we applied Bolasso as a feature selection method, and narrowed down the features suitable for interpretation of each shadow. Bolasso is feature selection method which is combination of Lasso and bootstrap method. This method estimates effective features from feature combination sets obtained by repeating data resampling and selecting features using Lasso . In the experiment, we used artificial data to show the effectiveness of Bolasso, and for lung CT images including diffuse lung disease, we estimated effective features for Interpretation and evaluated., Domestic conference
10 Jun. 2019

AIの基礎と展望
庄野 逸
Invited oral presentation, Japanese, 第117回 日本医学物理学会学術大会, Invited, 日本医学物理学会, パシフィコ横浜, http://www.jsmp.org/conf/117/, Domestic conference
14 Apr. 2019

VGGモデルの視覚野的解釈における解析の検討
寺元 陶冶; 庄野 逸
Oral presentation, Japanese, 電子情報通信学会 ニューロコンピューティング研究会, 電子情報通信学会, 電気通信大学, Domestic conference
05 Mar. 2019

辞書学習を用いたPET画像再構成
奥村 直裕; 庄野 逸
Oral presentation, Japanese, 電子情報通信学会 ニューロコンピューティング研究会, 電子情報通信学会, 電気通信大学, Domestic conference
05 Mar. 2019

MicroCT画像のための超解像とノイズ除去の検討
眞下 美紅; 庄野 逸
Oral presentation, Japanese, 電子情報通信学会 ニューロコンピューティング研究会, 電子情報通信学会, 電気通信大学, Domestic conference
05 Mar. 2019

テクスチャ画像識別問題に対するフーリエ変換を用いたデータ拡張の検討
新田 大悟; 庄野 逸
Oral presentation, Japanese, 電子情報通信学会 ニューロコンピューティング研究会, 電子情報通信学会, 電気通信大学, Domestic conference
05 Mar. 2019

SVCCAを用いた異なるデータセットで訓練されたDCNNの類似性測定
寺元 陶冶; 庄野 逸
Oral presentation, Japanese, 電子情報通信学会 ニューロコンピューティング研究会, 電子情報通信学会, 北海道大学, Domestic conference
Jan. 2019

ベイズ的変数選択に基づく分光スペクトル分解
川島 貴大; 庄野 逸
Oral presentation, Japanese, 数理モデル化と問題解決研究会, 情報処理学会, 電気通信大学, Domestic conference
Dec. 2018

特徴選択における状態探索手法の比較検討
遠藤 瑛泰; 永田 賢二; 庄野 逸
Oral presentation, Japanese, 情報処理学会数理モデル化問題解決研究会, 情報処理学会, 電気通信大学, Domestic conference
Dec. 2018

問題への適切生を考慮した畳み込みニューラルネットワークの初期値決定手法
鈴木 藍雅; 庄野 逸; 坂無 英徳
Oral presentation, Japanese, 数理モデル化と問題解決研究会, 情報処理学会, 小樽商科大学, Domestic conference
26 Sep. 2018

報科学屋さんからみたディープラーニング像
庄野 逸
Invited oral presentation, Japanese, 第46回 日本磁気共鳴医学会大会, Invited, 日本磁気共鳴医学会, Domestic conference
Sep. 2018

ディープラーニングを用いた医用画像識別の実現
庄野 逸
Invited oral presentation, Japanese, 千葉県非破壊検査研究会, 千葉県 非破壊検査研究会, Domestic conference
Jul. 2018

びまん性肺疾患診断における階層的特徴選択アプローチ
遠藤 瑛泰; 永田 賢二; 木戸 尚治; 庄野 逸
Oral presentation, Japanese, 情報処理学会 数理モデル化と問題解決研究会, 沖縄科学技術大学院大学, Domestic conference
15 Jun. 2018

ディープラーニングを用いた画像テクスチャ解析 データ駆動科学への橋渡しを目指して
庄野 逸
Invited oral presentation, Japanese, 第3回AIXセミナー, Invited, 電気通信大学 先端人工知能研究センター, Domestic conference
Jun. 2018

コンボリューションニューラルネットワークの基礎と画像信号処理への応用
庄野 逸
Invited oral presentation, Japanese, システム制御情報学会発表会, Invited, システム制御情報学会, 京都, Domestic conference
May 2018

ディープラーニングを用いた画像処理
庄野 逸
Invited oral presentation, Japanese, 2018年度情報処理学会 北陸支部総会, Invited, Domestic conference
May 2018

ディープラーニングの基礎
庄野 逸
Public discourse, Japanese, 第１９回情報論的学習システム(IBIS) 2016 チュートリアル, Invited, 京都, http://ibisml.org/ibis2016/tutorial-detail/, Domestic conference
19 Nov. 2017

ディープラーニングと画像処理への応用
庄野 逸
Public discourse, Japanese, 電気通信大学100週年記念行事スマートテクノロジーフォーラム2017, Invited, 電気通信大学, Domestic conference
27 Sep. 2017

Medical Texture Image Classification using Deep Convolution Neural Network with Transfer Style learning
Hayaru Shouno
Invited oral presentation, Japanese, 日本神経回路学会, Invited, 第27回日本神経回路学会全国大会, 北九州国際会議場, Domestic conference
22 Sep. 2017

Network in Network における Cascaded Cross Channel Pooling の解析
黒坂 衛; 庄野 逸
Poster presentation, Japanese, 第27回 日本神経回路学会全国大会, 日本神経回路学会, Domestic conference
21 Sep. 2017

Texture classification on Medical CT image using Deep Learning
Hayaru Shouno
Invited oral presentation, English, 第55回日本生物物理学会大会, Invited, 日本生物物理学会, Domestic conference
20 Sep. 2017

医用画像識別におけるスパース特徴選択手法について
庄野 逸
Invited oral presentation, Japanese, 電子情報通信学会ソサエティ大会, Invited, 電子情報通信学会, 東京都市大学, Domestic conference
12 Sep. 2017

ディープラーニングを用いた医用画像工学応用
庄野 逸
Invited oral presentation, Japanese, 第４２回日本光学会シンポジウム, Invited, 日本光学会, 東京大学生産技術研究所, Domestic conference
21 Jun. 2017

Deep Learning を用いた スパーステクスチャ画像解析手法の確立
庄野 逸
Oral presentation, Japanese, 科学研究費補助金新学術領域研究「スパースモデリングの深化と高次元データ駆動科学の創成」 2017 年度第 1 回公開シンポジウム, Domestic conference
07 Jun. 2017

AI and Deep Learning for Computer Aided Diagnosis: Present and Future
庄野 逸
Invited oral presentation, Japanese, 第73回日本放射線技術学会総会学術大会, Invited, 日本放射線技術学会, Domestic conference
15 Apr. 2017

AI and Deep learning for Computer Aided Diagnosis: Present and future
Hayaru Shouno
Invited oral presentation, Japanese, 第73回日本放射線技術学会総会学術大会, Invited, 日本放射線技術学会, パシフィコ横浜, Domestic conference
15 Apr. 2017

音楽の三要素からの生成モデルアプローチによる音楽生成手法の提案
川村誠護; 寺田英雄; 庄野 逸
Oral presentation, Japanese, 第79回情報処理学会全国大会, 名古屋大学, Domestic conference
16 Mar. 2017

階層型確率的主成分分析モデルによるテクスチャの生成
鈴木 藍雅; 庄野 逸
Oral presentation, Japanese, 電子情報通信学会 ニューロコンピューティング研究会, http://www.ieice.org/ken/paper/20170313Zbs2/, Domestic conference
13 Mar. 2017

Super-resolution with Deep Learning style
Hayaru Shouno; Yoshihiro Kusano
Poster presentation, English, The 3rd International Symposium on Multi-disciplinary Computational Anatomy, 部科学省科学研究費補助金新学術領域研究「医用画像に基づく計算解剖学の多元化と高度知能化診断・治療への展開」, Domestic conference
08 Mar. 2017

Applying Transfer Method for Deep Learning from Application Viewpoint
Hayaru Shouno
Invited oral presentation, English, International Forum on Medical Image in Asia (IFMIA) 2017, Okinawa, http://ifmia2017.may-pro.net/, International conference
17 Jan. 2017

ニューラルネットワークの基本と歴史
庄野 逸
Public discourse, Japanese, 日本神経回路学会 時限研究会「ニューラルネットの温故知新」, Invited, 日本神経回路学会 時限研究会「ニューラルネットの温故知新」, 電気通信大学, http://daemon.inf.uec.ac.jp/ja/events/ei4o9x/, Domestic conference
26 Sep. 2016

Network In Networkの視覚システムとしての妥当性について ～ 方位選択性マップに関する観点から ～
鈴木 聡志; 庄野 逸
Oral presentation, Japanese, 電子情報通信学会 情報論的学習理論研究会, Domestic conference
Sep. 2016

ディープラーニングの画像診断応用に向けて
庄野 逸
Invited oral presentation, Japanese, 電子情報通信学会 信号処理研究会, Invited, 電子情報通信学会 信号処理研究会, 千葉工業大学, Domestic conference
Sep. 2016

ディープラーニングの医用画像への応用
庄野 逸
Public discourse, Japanese, 日本画像医療システム工業会 医用画像システム部会, Invited, 日本画像医療システム工業会, http://www.jira-net.or.jp/commission/system/index.html, Domestic conference
Aug. 2016

Deep Neural Network の基礎
庄野 逸
Public discourse, Japanese, 第35回日本医用工学会大会, Invited, 日本医用画像工学会, 千葉大学, http://jamit2016.jamit.jp/, Domestic conference
21 Jul. 2016

ディープラーニングの医用画像への応用-人工知能時代へ向けて-
庄野 逸
Invited oral presentation, Japanese, 医用画像情報学会(MII) 平成28年度年次(第175回)大会, Invited, 医用画像情報学会, 大阪府立病院, http://mii-sci.sakura.ne.jp/wps/2016/05/15/, Domestic conference
15 May 2016

Deep Learning の理解と展望
庄野 逸
Oral presentation, Japanese, 電子情報通信学会 東海支部 専門講習会, Invited, 電子情報通信学会 東海支部, 名古屋キャッスルプラザ, http://www.ieice.org/tokai/general/specialty-lecture/, Domestic conference
29 Mar. 2016

ドロップアウトの正則効果に関する研究
近藤 佑; 鈴木 聡志; 斎藤 大輔; 原 一之; 庄野 逸
Oral presentation, Japanese, 第71回 日本物理学会春季大会, Domestic conference
16 Mar. 2016

Deep Convolution Netを用いたCT画像超解像の試み
Yoshihiro Kusano; Hayaru SHouno
Oral presentation, Japanese, 電子情報通信学会 ニューロコンピューティング研究会, Domestic conference
Mar. 2016

びまん性肺疾患陰影の Deep Neural Network による識別手法の構築
庄野 逸
Invited oral presentation, Japanese, 第10回次世代コンピュータ支援診断ソフトウェア臨床使用・評価プラットフォーム研究会, Invited, http://www.ut-radiology.umin.jp/ical/CIRCUS/events/201512_10th_seminar/index.html, Domestic conference
19 Dec. 2015

Deep Learningによる画像処理
庄野 逸
Invited oral presentation, Japanese, 日本光学会 コンテンポラリーオプティクス研究グループ, Invited, 日本光学会 コンテンポラリーオプティクス研究グループ, https://docs.google.com/forms/d/1Q3c0wQnj12fZYVaxqWfdA0b4_90-2Bn3hsKjvDssgXw/viewform?c=0&w=1, Domestic conference
27 Nov. 2015

Novel texture classification with Deep Convolution Neural Network-Evaluation with Lung CT Images-
Hayaru Shouno
Invited oral presentation, English, International Symposium on Object Vision in Human, Monkey, and Machine, Invited, 電気通信大学，RIKEN BSI, UEC BLSC, 神経回路学会, University of Electro-Communications, http://www.cns.mi.uec.ac.jp/ovs/, International conference
06 Nov. 2015

MCMC法を用いたびまん性肺疾患画像の特徴量選択
Makoto Koiwai; Masaki Isogai; Hayaru Shouno; Shoji Kido
Oral presentation, Japanese, 電子情報通信学会 医用画像研究会, 電子情報通信学会 医用画像研究会, 東京, Domestic conference
08 Sep. 2015

MCMC法を用いたびまん性肺疾患画像の特徴量選択
小岩井誠; 磯谷真希; 庄野逸; 木戸尚治
Poster presentation, Japanese, 第25回日本神経回路学会全国大会, 日本神経回路学会, Domestic conference
02 Sep. 2015

ネオコグニトロンにおける識別率と細胞反応分布の関係について
佐藤翔一郎; 菊池眞之; 福島邦彦; 林勲,庄野逸
Poster presentation, Japanese, 第25回日本神経回路学会全国大会, 日本神経回路学会, Domestic conference
02 Sep. 2015

Deep Convolutional Neural Network の特徴抽出に関する特性ーびまん性肺疾患を例にしてー
鈴木聡志; 庄野逸; 木戸尚治
Poster presentation, Japanese, 第25回日本神経回路学会全国大会, 日本神経回路学会, Domestic conference
02 Sep. 2015

びまん性肺疾患識別用DCNNの階層毎の特徴解析
Satoshi Suzuki; Hayaru Shouno; Shoji Kido
Nominated symposium, Japanese, The 18th Meeting on Image Recognition and Understanding, Domestic conference
30 Jul. 2015

Diffuse Lung Disease Pattern Recognition with Deep Convolutional Neural Network
Hayaru Shouno; Satoshi Suzuki; Shoji Kido
Nominated symposium, English, 11th Asia-Pacific Conference on VIsion, Invited, Singapore, http://apcv2015.org, International conference
11 Jul. 2015

Anaysis for Deep Convolutional Neural Network feature with Diffuse Lung Disease classification
Satoshi Suzuki; Hayaru Shouno; Shoji Kido
Oral presentation, Japanese, 情報処理学会 数理モデル化と問題解決研究会, 情報処理学会 数理モデル化と問題解決研究会, 沖縄科学技術大学院大学, http://id.nii.ac.jp/1001/00142366/, Domestic conference
16 Jun. 2015

Deep Convolutional Neural Networkを用いたびまん性肺疾患画像の特徴解析
Satoshi Suzuki; Hayaru Shouno; Shoji Kido
Oral presentation, Japanese, 電子情報通信学会 ニューロコンピューティング研究会, Domestic conference
16 Mar. 2015

ディープラーニング：階層型Neural Netの温故知新
庄野 逸
Invited oral presentation, Japanese, 電子情報通信学会 東海支部 一般講演会, Invited, 電子情報通信学会 東海支部, Domestic conference
08 Dec. 2014

階層型HOGにおける高次特徴抽出量のPCA解析
Naoki Yoshitake; Hayaru Shouno
Oral presentation, Japanese, 情報処理学会 MPS 研究会, Domestic conference
18 Sep. 2014

Transductive Support Vector Machineを用いたびまん性肺疾患画像の認識
Y. Hayakawa; H. Shouno; S.Kido
Oral presentation, Japanese, 電子情報通信学会技術報告,MI研究会
Oct. 2012

Classification of Idiopathic Interstitial Pneumonias using Semi-Supervised Learning
Masayoshi Wada; Hayaru Shouno; Shoji Kido
Oral presentation, Japanese, IEICE
Jul. 2012

A hierarchical extension of the HOG model implemented in the convolution-net for human detection
Y. Arakaki; H.Shouno; K.Takahashi; T.Morie
Oral presentation, English, 情報処理学会,数理モデル化と問題解決研究会
Mar. 2012

階層モデルを用いたMT野神経回路モデル
奈良紗友里; 庄野逸
Oral presentation, Japanese, 電子情報通信学会,ニューロコンピューティング研究会
Mar. 2012

Poissonノイズ画像に対する局所変分法を用いた画像修復
庄野逸; 瀧山健; 岡田真人
Oral presentation, Japanese, 日本物理学会,春季大会
Mar. 2012

A deterministic algorithm for hyperparameter estimation in nonlinear Markov random field model
OHNO Yoshinori; NAGATA Kenji; SHOUNO Hayaru; OKADA Masato
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/023346977, Image restoration widely used in natural science is often formulated by Markov random field (MRF) model. In MRF model, the framework of Bayesian inference allows us to estimate the hyperparameters through free-energy minimization. When a nonlinear function represents the observation process, Markov chain Monte Carlo method has been applied to hyperparameter estimation so far. While the previous method retains nonlinearity, it is a probablistic algorithm, which makes convergence judgement difficult and increases computational complexity. In this paper, we propose a deterministic algorithm for hyperparameter estimation in nonlinear MRF model, by linearizing the observation process to calculate the free-energy and estimate the hyperparameters analytically. Besides, we use artificial image data to show the efficiency of the proposed method.
17 Nov. 2011
17 Nov. 2011- 17 Nov. 2011

An image restoration method for Poisson observation using a latent variational approximation
SHOUNO Hayaru; TAKIYAMA Ken; OKADA Masato
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/023346949, In this study, we treat an image restoration problem throughout a Poisson noise channel observation. The Poisson noise channel is often hard to tract in a theoretical analysis because of its non-negative property. In our formulation, we interpret the Poisson noise channel observation as a Bernoulli process, and apply a latent variable method to transform the observation as a Gaussian process with single latent variable. We formulate the image restoration problem as a Bayesian approach, and introduce a Gaussian Markov random field for its prior. The latent parameters and hyper-parameter, which determine the balance ratio between the observation fidelity and prior knowledge, are estimated by maximization of marginalized log-likelihood, known as "evidence" or "free energy". In order to maximize marginalized log-likelihood, we introduce an expectation maximization algorithm.
17 Nov. 2011
17 Nov. 2011- 17 Nov. 2011

A latent variational approximation method of Total variation for noise reduction
SHOUNO Hayaru; OKADA Masato
Japanese, 電子情報通信学会技術研究報告 : 信学技報, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/023346071, In the field of noise reduction for the observed signal, the total variation is one of a standard constraint. The total variation, which prefer the small diference between neighbor units, is discussed in a context of LI-norm minimization. In this study, we treat it as a Laplace distribution, and approximate it by a latent variational method. The latent variational method translate the distribution as a Gaussian distribution form with a latent parameter. We propose a method applying the expectation-maximization algorithm to estimate both the latent parameters and hyper-parameters as a hidden variables.
09 Nov. 2011
09 Nov. 2011- 09 Nov. 2011

局所変分法を用いた Total Variation の近似とノイズ除去
庄野逸; 岡田真人
Oral presentation, Japanese, 第14回情報論的学習理論ワークショップ(IBIS2011),第14回情報論的学習理論ワークショップ(IBIS2011)
Nov. 2011

非線形マルコフ確率場モデルのハイパーパラメータ推定における決定的アルゴリズム
大野義典; 永田賢二; 庄野逸; 岡田真人
Oral presentation, Japanese, 電子情報通信学会,ニューロコンピューティング研究会
Nov. 2011

Classification of Idiopathic Interstitial Pneumonia CT Images using Convolutional-net with Sparse Feature Extractors
Taiju Inagaki; Hayaru Shouno; Shoji Kido
English, 研究報告数理モデル化と問題解決（MPS）, http://id.nii.ac.jp/1001/00075149/, We propose a computer aided diagnosis (CAD) system for classification of idiopathic interstitial pneumonias (IIPs). High resolution computed tomography (HRCT) images are considered as effective for diagnosis of IIPs. Our proposed CAD system is based on the convolutional-net that is bio-plausible neural network model inspired from the visual system such like human. The convolutional-net extract local features and integrate them in the process of hierarchical neural network system. For natural image recognition by convolutionalnet, Gabor feature extraction is known to give a good performance, however, the HRCT images may have different properties from those of natural images. Thus, we introduce a learning type feature extraction called "sparse coding" into the convolutional-net, and evaluate performance for classification of IIPs.We propose a computer aided diagnosis (CAD) system for classification of idiopathic interstitial pneumonias (IIPs). High resolution computed tomography (HRCT) images are considered as effective for diagnosis of IIPs. Our proposed CAD system is based on the convolutional-net that is bio-plausible neural network model inspired from the visual system such like human. The convolutional-net extract local features and integrate them in the process of hierarchical neural network system. For natural image recognition by convolutionalnet, Gabor feature extraction is known to give a good performance, however, the HRCT images may have different properties from those of natural images. Thus, we introduce a learning type feature extraction called "sparse coding" into the convolutional-net, and evaluate performance for classification of IIPs.
11 Jul. 2011
11 Jul. 2011- 11 Jul. 2011

Bayesian Reconstruction for Computed Tomography using 3 Dimensional Markov Random Field
S. Kawato; M. Yamasaki; H.Shouno; M.Okada
Oral presentation, Japanese, 電子情報通信学会技術報告,IE2011-17, PRMU2011-9, MI2011-9
Jun. 2011

Bayesian Reconstruction for Computed Tomography using 3 Dimensional Markov Random Field
YAMASAKI Madomi; KAWATO Shohei; SHOUNO Hayaru; OKADA Masato
Japanese, IEICE technical report, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/11116791, We propose a 3-dimensional Bayesian CT reconstruction method in which we introduce 3-dimension Markov random field (MRF) like prior. In the Bayesian inference, the prior, which plays a role of compensation for degraded information, is defined as the property of the ideal image you consider. The target object is exist in the 3 dimensional world, so that it is natural to introduce such 3D MRF like prior. In this study, we evaluate the 3D Bayesian reconstruction method in the computer simulation.
12 May 2011
12 May 2011- 12 May 2011

Image restoration for computed tomography image with Bayesian approach
UEKI Junta; SHOUNO Hayaru
Japanese, IEICE technical report, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/11047297, When a doctor diagnose patient, a doctor use a computed tomography image. When we get computed tomography image, we expect low invasive image. However, when we take a computed tomography image, the image is debased by various problems. For instance, this problems occur by mechanical factors or by radiation irradiance of X-ray computed tomography scanner. Thus when we try to solve the problems, we can think the compensation mechanism of restoration process. On the other hand, we can think the directly approach that restore the computed tomography image, too. We focused on the approach of the latter and weigh the former against the latter. This approach has Bayes estimation for image restoration. We try to use belief propagation for this approach. This approach use prior distribution as image character and deterioration process as image degradation. We use prior distribution as Gaussian Markov random field and deterioration process as gaussian noize. Moreover, we estimate those Hyperparameter by maximizing a marginal likelihood. We weigh the image generated by this image restoration approach against the other image generated by the other image restoration approach that is the compensation mechanism of restoration process.
28 Feb. 2011
28 Feb. 2011- 28 Feb. 2011

An Improvement of Convolutional net by using of SIFT algorithm
ARAKAKI Yasuto; SHOUNO Hayaru
Japanese, IEICE technical report, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/11047441, Image recognition technology has been used in various fields. However, The recognition of the non-constant image such as handwriting image is still difficult. The solution to such problems seems to be referring to the architecture of visual processing in biology such as Lowe's model, Convolutional Net. Lowe constructed model to combin the architecture of visual processing and template matching method. However, Lowe's model include random sampling. So, the recognition rate was not stable. This report purposed to obtain the constant recognition rate. For this, We got in SIFT so that the randomness can be avoided in the system. We demonstrated recognition rate by using The MNIST database of handwritten digits, draw a comparison between The Lowe's model and The improved model.
28 Feb. 2011
28 Feb. 2011- 28 Feb. 2011

Improvement of infant's action recognition accuracy by Bayesian estimation method introducing Bayesian Network : Experimental evaluation with supersonic sensor and camera images
ISHIKAWA Shouzou; MOTOMURA Yoichi; NISHIDA Yoshifumi; SHOUNO Hayaru
Japanese, IEICE technical report, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/11046674, The purpose of this study is to prevent accident in infants. Therefore, we consider analysis the action of the behavior in the everyday life from several types of sensors. Conventional action recognition has been done only from the image. WWe propose applying additional information, which we treat as the prior distribution in the meaning of the Bayes inference, observed from the supersonic sensors. Prior distribution use Bayesian network formulation in the observation data. Likelihood function calculates maximum likelihood estimation method in feature extraction of images. In this paper, we consider feature extraction candidates as HLAC, SIFT, and 3D SIFT, and compare the performance of them. We estimate behavior labels by this method. Then, we performed a comparison experiment to inference a behavior labels by this method.
28 Feb. 2011
28 Feb. 2011- 28 Feb. 2011

局所変分法を用いた Poisson 過程観測下における画像修復
庄野逸; 瀧山健; 岡田真人
Oral presentation, Japanese, 電子情報通信学会,ニューロコンピューティング研究会
2011

Classification of Idiopathic Interstitial Pneumonia on High-resolution CT Images using Counter-propagation Network
TANAKA YUKI; SHOUNO HAYARU; KIDO SHOJI
English, 研究報告数理モデル化と問題解決（MPS）, 情報処理学会, http://id.ndl.go.jp/bib/025047781, In order to classify the idiopathic interstitial pneumonias(IIPs), extraction and interpretation of features on high-resolution computed tomography (HRCT) image is considered to be effective. The purpose of our study is to develop a diagnosis support system to help diagnostician of classification for those HRCT images using an artificial neural network called counter propagation network. The CPN is a hybrid type neural network model composed from self-organizing map (SOM) for feature extraction and from multi-layered perceptron (MLP) for classification. Applying the CPN for the IIPs images, we could obtain both a kind of similarity map and classification system.In order to classify the idiopathic interstitial pneumonias(IIPs), extraction and interpretation of features on high-resolution computed tomography (HRCT) image is considered to be effective. The purpose of our study is to develop a diagnosis support system to help diagnostician of classification for those HRCT images using an artificial neural network called counter propagation network. The CPN is a hybrid type neural network model composed from self-organizing map (SOM) for feature extraction and from multi-layered perceptron (MLP) for classification. Applying the CPN for the IIPs images, we could obtain both a kind of similarity map and classification system.
05 Jul. 2010
05 Jul. 2010- 05 Jul. 2010

Neocognitron Trained by a New Competitive Learning
FUKUSHIMA Kunihiko; HAYASHI Isao; SHOUNO Hayaru; KIKUCHI Masayuki; MAKINO Yuki
Japanese, IEICE technical report, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/10649843, The neocognitron is a hierarchical multilayered neural network capable of robust visual pattern recognition. It acquires the ability to recognize patterns through learning. This paper proposes several improvements made on the neocognitron: new competitive learning method with winner-kill-loser rule, dis-inhibition to the inhibitory surround of the receptive fields of C-cells (or complex cells), square-root shaped non-linearity in the input-to-output characteristics of C-cells, and so on. As a result of these improvements, the recognition rate of the neocognitron has been largely increased. We also reduced the number of parameters that have to be determined in designing a neocognitron.
02 Mar. 2010
02 Mar. 2010- 02 Mar. 2010

Edge Extraction for the Neocognitron
MAKINO Yuki; KIKUCHI Masayuki; FUKUSHIMA Kunihiko; HAYASHI Isao; SHOUNO Hayaru
Japanese, IEICE technical report, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/10649847, Neural network model neocognitron has an ability of robust visual pattern recognition. Feature-extracting cells, called S-cells, in the first stage of the neocognitron, extract oriented edges from input patterns. Under certain conditions, however, t he S-cells yield spurious outputs in the places where edges do not actually exist. Occurrence of the spurious edge is might lower the recognition rate. To suppress spurious responses, in the conventional neocognitron, an inhibitory surround is introduced in the input connections to C-cells, which make a blurring operation. We proposed a new method of edge extraction that suppress spurious edges, and tested how the recognition rate of the neocognitron is improved. As a result, spurious responses have been suppressed, but there was not a large difference in the recognition rate.
02 Mar. 2010
02 Mar. 2010- 02 Mar. 2010

Image Reconstruction for Radon Transform using Bayes Inference
SHOUNO Hayaru; OKADA Masato
Japanese, IEICE technical report, The Institute of Electronics, Information and Communication Engineers, http://ci.nii.ac.jp/naid/110007866389, We propose an image reconstruction algorithm using Bayesian inference for the Radon transformed observation data, which is often appeared in the field of medical image reconstruction problem known as computed tomography(CT). In order to apply our Bayesian reconstruction method, we introduced several hyper-parameters which controls a kind of the ratio between prior information and a fidelity of the observation process. Since the quality of the reconstructed image is influenced by the estimation accuracy of those hyper-parameters, we propose an inference method for these hyper-parameters based on the marginal likelihood maximization principle as well as the method for image reconstruction. We show an improved reconstruction result rather than that of a conventional method, which is called Filtered Back Projection(FBP).
21 Jan. 2010
21 Jan. 2010- 21 Jan. 2010

Bayesian Inference in combination with Tree Augmented Naive Bayes and Bayesian Networks for Infant's Behavior Recognition accuracy comparing
ISHIKAWA SHOZO; MOTOMURA YOICHI; NISHIDA YOSHIFUMI; SHOUNO HAYARU
Japanese, 研究報告バイオ情報学（BIO）, 情報処理学会, http://id.ndl.go.jp/bib/025071209, 近年，センサ技術の発達によって人間の行動を観測することが可能になり，各種センサから大量のデータを取得することが容易になっている．しかし，このセンサデータと行動の意図を結びつけることは容易ではない．そこで，この大量のセンサデータに対して行動ラベルを割り当てることを考える．大量のセンサデータに対して手作業でラベルを割り当てることは困難なのでセンサデータから行動ラベルを推定することを考える．本稿では幼児の事故予防を目的とし，幼児の室内行動に注目して，幼児の行動を記録した動画と超音波センサデータから行動ラベルを推定する．推定の手法にはセンサデータから構築したベイジアンネットワークと TAN 識別器 (Tree Augmented NaiveBayes Classifier) を統合したベイズ推定を使用する．ベイズ推定においては，ベイジアンネットワークの出力を事前分布として用い，尤度関数は，動画像における高次局所自己相関特徴量をTAN識別器にかけた出力を用いる．以上のようなベイズ推定を用い行動ラベル推定の比較実験を行った．The purpose of this study is to prevent accident in Infants. Recent years, we can observe human's behavior by gaining sensor technology. We can get easily a lot of observation data from a wide variety of sensors. However, It is not easy for tying this observation data to the intention of the human's behavior. So, we consider behavior label is allocated to a lot of observation data. We pay attention to the infant's indoor behavior in this paper. We estimate behavior labels from observation data from ultra sonic sensors and a fisheye camera . Inference method is using Bayes inference by combination with Tree Augmented Naive Bayes(TAN) and Bayesian Networks. These models are constructed from observationdata. Prior distribution is using Bayesian Networks. Likelihood function is using Tree Augmented Naive Bayes Classifier which is constructed from higher order local autocorrelation features from images in infant surrounding. We estimate behavior labels by this method. Then, we performed a comparison experiment to inference an behavior labels by this method.
10 Dec. 2009
10 Dec. 2009- 10 Dec. 2009

Feature extraction with Sparse-coding for medical images
INAGAKI TAIJU; SHOUNO HAYARU
Japanese, 研究報告バイオ情報学（BIO）, http://id.nii.ac.jp/1001/00067020/, コンピュータ支援診断 (computer-aided diagnosis, CAD) において，CT 画像などの画像診断を行うようなシステムを構築する場合，画像クラス分類は重要な課題である．一般に，医師が経験で得た知識をアルゴリズムとして定義することは難しいため，計算機で実装することは困難を伴う場合が多い．本研究は，ヒトが画像の判別を行うようなメカニズムを何らかの形で CAD システムに取り入れることで，このクラス分類に対するアプローチを行うことを目的としている．ここでは，クラス分類において重要なウェイトを占める特徴抽出に対して，生物の視覚の学習モデルとして考案されたスパースコーディングを適用し，画像データから抽出された特徴が有効なものであるかの評価を試みた．Image classification is a important problem when we build a system preforming the image diagnosis such as CT in Computer-Aided Diagnosis(CAD). In general, it has many cases with the difficulty to implement with a computer, because it is difficult to define the knowledge that a doctor got by experience as algorithm. In this sutdy, there is it for the purpose of performing approach for this class classification taking the mechanism that the human distinguishes the image into CAD. And, we tried the evaluation that feature quantity extracted by sparse coding which devised as a learning model of the sighy of the human from image data was effective.
10 Dec. 2009
10 Dec. 2009- 10 Dec. 2009

Reconstruction of tomographic image based on Bayes approach
YAMAMOTO SHO; SHOUNO HAYARU
Japanese, 研究報告バイオ情報学（BIO）, http://id.nii.ac.jp/1001/00067022/, 観測データから画像を再構成する技術は，様々な分野で非破壊検査の手法として応用されてきている．このうち信頼性の高い観測データが得られる場合では，最尤推定法に基づく手法が有効であるとされ，医療分野では ML-EM 法といった手法が用いられるようになっている．しかし観測データが低品質な場合，信号がノイズに埋もれてしまい，うまく再構成できない場合がある．ノイズに埋もれたデータからでも高画質に画像を再構成する方法として，ベイズアプローチを用いた MAP-EM 法が注目されている．本研究では ML-EM 法，MAP-EM 法それぞれの手法で画像再構成の計算機シミュレーションを行い，また MAP-EM 法についての改善点について検討した．The technology to reconstruct an image from observation data has been applied as technique of nondestructive inspection in various fields. When we could get reliable observation data, method based on Maximum likelihood estimation is effective, and ML-EM method comes to be used in medical field. However, when observation data were low quality, A signal is buried among noises, and there is the case that I cannot reconstruct well. MAP-EM method which used idea based on Bayes approach is interested because we may reconstruct high quality image from such data. We simulate reconstruction method ML-EM and MAP-EM and consider improvement of MAP-EM in this paper.
10 Dec. 2009
10 Dec. 2009- 10 Dec. 2009

Image Reconstruction for Medical Image using Bayes approach
H.Shouno; M.Okada
Invited oral presentation, Japanese, 情報統計力学の広がり：量子・画像・そして展開, Depending and Expansion of Statistical Mechanical Infomatics
Jul. 2009

Extraction of Pleural Effusion Regions from 3-D Thoracic CT Images
TSUNOMORI Akinori; SHOUNO Hayaru; KIDO Shoji
Japanese, IEICE technical report, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/9378135, Computed tomography(CT) is used to diagnose pleural effusion. It is an effective method for follow-up to measure accumulating volume of pleural effusion quantitatively. But, it is difficult for doctors to measure them quantitatively from 3-D thoracic CT images. So, a method for segmentation of pleural effusion is needed. We used that CT values are homogeneous in pleural effusion regions and pleural effusion regions locate at back side in the supine position. To evaluate our developing method, we compare the three kinds of extraction results: the first is the manually extracted result, the second is a simple method based on the threshold method, and the third is our extraction method. As a result, we confirmed our method is effective for the case with pleural effusion.
25 Jan. 2008
25 Jan. 2008- 25 Jan. 2008

Consideration of a method for evaluating airway wall with 3D-thoracic CT images
MIYAMOTO Satoshi; SHOUNO Hayaru; KIDO Shoji
Japanese, IEICE technical report, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/9378014, Evaluating airway wall is an important factor for diagnosis of COPD(Chronic Obstructive Pulumonary Disease) by use of CT (computed tomography) images. COPD is a disease that changes airway wall thickness, and a diagnosis with evaluating airway wall for it is carried out by a doctor manually by use of each two-dimensional CT slice image. However, we consider the three-dimensional evaluation of airway wall thickness is more effective for accurate COPD diagnosis. And that airway wall is consisted of two factors; one is airway wall area (outer wall area) and the other is inside area (inner wall area). So we must detect outer wall area and inner wall area to evaluate them automatically. In this paper, we propose a method for analyzing air area (inner wall area) and wall area (outer wall area). Especially, there are unwanted objects like vessels around the outer wall in clinical three dimensional thoracic CT images. So we propose a evaluating method to analyze them and checked a method by use of phantom data.
25 Jan. 2008
25 Jan. 2008- 25 Jan. 2008

Construction of a Supporting System for Portable Terminal Devices using Middle-ware Server to Communicate with DICOM Service
Seto Yuichi; Kido Shoji; Shouno Hayaru
Journal of Life Support Engineering, The Society of Life Support Engineering, https://jlc.jst.go.jp/DN/JALC/00370472537?from=CiNii
2008
2008 2008

Anaysis of Idiopathic interstitial Pneumonia by Self Organization Map on High-resolution Computed Tomography Images
GOTO Yoshiharu; SHOUNO Hayaru; KIDO Shoji
English, IPSJ SIG Notes, Information Processing Society of Japan (IPSJ), http://id.ndl.go.jp/bib/8803371, In classifying the idiopathic interstitial pneumonias (IIPs), interpretation of features on high-resolution computed tomography (HRCT) image is effective. However, image patterns of IIPs on HRCT images have so much variety, that the classifying problem is difficult. The purpose of our study is to develop a diagnosis support system for classification of those HRCT images using a Kohonen's self-organizing map (SOM). Our system classify the input HRCT image as 4 IIP classes, that is, Consolidation, Ground-Grass, Honeycomb, and Reticular classes.
25 Jun. 2007
25 Jun. 2007- 25 Jun. 2007

Detection algorithm of small hepatic tumors by use of multi-phase multi-detector row CT imagesa
MIYAZAKI Hiroyasu; SHOUNO Hayaru; KIDO Shoji
Japanese, IEICE technical report, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/8638743, For the recent tequnical advance of multidetector-row computed tomography (MDCT), which enables to scan a human body quickly and minutely, hepatic images including small and faint hepatic tumors that are difficult to find out by conventional CT images. However, the detection of these lesions requires radiologists' carefulness even by use of MDCT images, and such task burdens radiologists. In this study, we propose a method for detecting hepatic tumors using both contrast-enhanced (CE) and non-CE CT images. In our method, at first, we apply phase only correlation method for registration between CE and non-CE CT images. After that, we subtract non-CE images from CE images, and obtain subtracted images which include candidates of small and faint hepatic tumors. The subtracted images, however, include many false positives (FPs), and these FPs caused by blood vessels in the liver. Thus, in our algorithm, FPs based on vessels are reduced from tumor candidates.
20 Jan. 2007
20 Jan. 2007- 20 Jan. 2007

Development of Computerized System for Selection of Similar Images from Different Patients for Image Subtraction of Chest Radiographs
ODA Nobuhiro; AOKI Takatoshi; OKAZAKI Hiroko; KAKEDA Shingo; KOUROGI Yukunori; YAHARA Katuya; SHOUNO Hayaru
Japanese, Transactions of the Japanese Society for Medical and Biological Engineering : BME, Japanese Society for Medical and Biological Engineering, http://id.ndl.go.jp/bib/8613681, The purpose of this study was to develop a novel computerized scheme to automatically select similar chest radiographs for image subtraction of patients who have no previous chest radiographs and to assist the radiologist's interpretation by presenting "similar subtraction images". A large database of approximately 15,000 posteroanterior chest radiographs, which were diagnosed as normal, was used for searching similar images of different patients. First, in this scheme, two clinical parameters (age and sex) were used for selecting similar images. Next, 100 images of candidates in the database were selected according to similarity in height and the area of the lung field in the target image. We used quantitative measurement for searching similar images; namely, the correlation value of cheat region in the 100 images of the candidates. The similar subtraction images were obtained by subtracting the similar images selected from the target image. The performance of the proposed system was evaluated in comparison with 95 chest radiographs with a temporal subtraction image. The experimental results showed that the average of the correlation values in the temporal subtraction image and similar subtraction images were 0.9794 and 0.9574, respectively. Three radiologists subjectively evaluated various lung artifacts on the temporal subtraction image and similar subtraction images of 95 chest radiographs using a five-point rating scale (1:very poor, 2:poor, 3:adequate, 4:good, 5:excellent). Ratings higher than "adequate" were given for 70% of the similar subtraction images. This computerized scheme seems useful for the automatic selection of similar images for similar-image subtraction of chest radiographs and has potential use for assisting interpretations by radiologists.
10 Sep. 2006
10 Sep. 2006- 10 Sep. 2006

Discrimination of Lung Sounds using a Statistics of Waveform Intervals
ORIHASHI Taketoshi; SHOUNO Hayaru; KIDO Shoji
English, IPSJ SIG Notes, Information Processing Society of Japan (IPSJ), http://id.ndl.go.jp/bib/7971915, In a lung auscultatory sounds diagnosis, the diagnosis result would be affected by the skill of the doctor, that is, the doctor should discriminate a lung sounds by own subjective, since standard diagnosis procedure with objectivity has not been established yet. In many cases of the lung sounds diagnosis, the existence of features, which are called adventitious sounds, are important key. The adventitious sounds is roughly classified into two class; one is called coarse crackles, and the other is called fine crackles. Thus, we construct a computer aided diagnosis (CAD) system for classifying a lung sound into three types, that is, coarse crackle, fine crackle and normal breath sound. We aimed with the waveform time intervals for discrimination. Our CAD system calculates average histograms of intervals for each class, and discriminates a input histogram into these three types by the distance between each average histogram.
26 Jun. 2006
26 Jun. 2006- 26 Jun. 2006

びまん性肺疾患の知的CAD
木戸 尚治; 四方 秀則; 庄野 逸; 松本 常男; 松永 尚文
Japanese, 日本医学放射線学会学術集会抄録集, (公社)日本医学放射線学会
Feb. 2006
Feb. 2006 Feb. 2006

Bronchus Extraction from Three Dimensional Chest CT Images for COPD Diagnosis
TAMECHIKA Shinya; SHIKATA Hidenori; SHOUNO Hayaru; KIDO Shoji
Japanese, IEICE technical report, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/7814789, Measurement of bronchus diameter is an important factor for diagnosis of COPD (Chornic Obstructive Pulmornary Disease) by use of computed tomography (CT) images. The measurement can be manually carried out for each two-dimensional CT slice by a doctor practically, however, we consider the three-dimensional (3D) measurement of bronchus diameter is more effective for accurate COPD diagnosis. This 3D measurement requires accurate segmentation of trachea and bronchus regions, however, application of conventional naive threshold-based method does not give enough extraction result because of a partial volume effect. In this paper, we propose an extraction algorithm, which is based on a local and adaptive region-growing method starting from trachea region, and the algorithm was applied to several cases. In addition, in order to perform the local region-growing method, a local seed point needs to be specified for each local region, so that, we introduce an algorithm for detecting seed points automatically.
20 Jan. 2006
20 Jan. 2006- 20 Jan. 2006

Medical treatments classification using PMM
IMAMRUA Yuichi; KIDO Shoji; SHOUNO Hayaru
Japanese, IPSJ SIG Notes, Information Processing Society of Japan (IPSJ), http://id.ndl.go.jp/bib/7765243, In recent years, information mining technique for text data becomes hot topic, especially for data in the Internet web site, and a lot of methods are developed. On the other hand, a lot of clinical record becomes to be saved with computerized media in a lot medical facilities. In this research, to investigate whether those textmining techniques are efficient for such medical treatment data, we have developed a classification application using the parametric mixture model (PMM). PMM, that assumes text data consists of "bag of words" and belongs to several given topics, infers which topics the text data are belonged to. Our medical treatment data, which describes about diseases of lungs, can be regarded as 3 types of topics, that is, "tumor", "inflammation", and "the other". We tried to classify those medical treatment data.
20 Dec. 2005
20 Dec. 2005- 20 Dec. 2005

Statistical Mechanics of a Spike Analysis Method using Log-Linear Model
SHOUNO Hayaru; OKADA Masato
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, http://ci.nii.ac.jp/naid/10016576474, Recently, Multi-electrodes detector, which consists of a lot of electrodes, is enable to measure a lot of neuron spike signals simultaneously. To analyze the spike signals, we applied a log-linear model. In this study, we analyze the log-linear model of spike signals as a Ising-spin system, which is used in the field of statistical mechanics. In the log-linear model, we assumed that the interactions in the model are obeyed to the multidimensional normal distribution, and the noise flip is occurred in the observed process. To estimate the interactions, which is called hyper-parameters, we use maximization of log-likelihood function. For maximization process, we applied gradient accent method and EM algorithm, and derived averaged evaluations for each algorithm.
16 Jun. 2005
16 Jun. 2005- 16 Jun. 2005

14aTD-12 Statistical Mechanics of a Spike Analysis Method with Log-Linear Model
Shouno H; Wada K; Okada M
Japanese, Meeting abstracts of the Physical Society of Japan, The Physical Society of Japan (JPS)
25 Aug. 2004
25 Aug. 2004- 25 Aug. 2004

Statistical Mechanics of a Spike Analysis Method with Log-Linear Model
SHOUNO Hayaru; WADA Koji; OKADA Masato
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/7036813, Recently, Multi-electrodes, which have a lot of electrode sensors, is enable to measure a lot of neuron spike signals. To analyze the spike signals, a log-linear model is tried to applied. In this study, we analyze the log-linear model of spike signals as a Ising-spin system, which is used in the field of statistical mechanics. In the log-linear model, we assumed that the interactions of each signal are uniform, and the noise flip is occurred in the observed process. To estimate the interactions, which is called hyper-parameters, we use maximization of log-likelihood function. For maximization process, we applied gradient accent method and EM algorithm, and derived averaged evaluations for each algorithm.
17 Jun. 2004
17 Jun. 2004- 17 Jun. 2004

30aWB-6 Correctness of Bethe approximation in probabilistic image processing by means of Gaussian graphical model
Tanaka Kazuyuki; Shouno Hayaru; Okada Masato
Japanese, Meeting abstracts of the Physical Society of Japan, The Physical Society of Japan (JPS)
03 Mar. 2004
03 Mar. 2004- 03 Mar. 2004

Naive Mean Field Approximation for Error Correcting Code Problem
TAKATA Masami; SHOUNO Hayaru; OKADA Masato; JOE Kazuki
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/6554936, In the communication theory, solving error correcting code problem is a important subject. Recently, to reveal the charactericity of the erroro correcting code, several reseachers applied statistical-mechanical approach to this problem. In our research, we treated the error correcting code as the Bayes inference framework. We used MPM (Maximizer of the Posterior Marginals) inference, that is a kind of Bayes inference, and, moreover, applied naive mean field approximation. In the field of the artificial neural network, this approximation is applied for reduction of the computational cost by substitution of the stochastic binary units with the deteministic continuousu value units. However, there exsists few report of the performance of this approximation by quantitatively. Therefore, we evaluated the performance of this approximation by analysis and compare it with the compuater simulation.
11 Mar. 2003
11 Mar. 2003- 11 Mar. 2003

Extraction of Train Noise affects Telluric Current Data at Two Different Observation Points by ICA and Its Statistical Evaluation
SAWA Sayuri; KOGANEYAMA Mika; SHOUNO Hayaru; NAGAO Toshiyasu; JOE Kazuki
Japanese, IPSJ SIG Notes, Information Processing Society of Japan (IPSJ), http://id.ndl.go.jp/bib/6381053, The method of detecting irregular changes of electric current in Telluric Current Data (TCD) has attracted notice recently as an effective method for short-term earthquake prediction. We call such irregular changes Seismic Electric Signals (SESs), which are sometimes observed before earthquake. However, since most of the TCD collected in Japan is affected by train noise, detecting SESs in TCD itself is considered as an extremely arduous job. The purpose of our research is to separate train noise and SESs automatically by ICA. Train noise and the SES are considered as independent source signals. It has been confirmed by our experiments that train noise observed at Matsushiro, Nagano can be separated by ICA. In this paper, we apply ICA to TCD observed at two different observation points which seems to contain the same train noise and evaluate the results by a statistical method.
28 Nov. 2002
28 Nov. 2002- 28 Nov. 2002

Analysis of Bidirectional Associative Memory using SCSNA and Statistical Neurodynamics
SHOUNO H.; OKADA M.
Japanese, IPSJ SIG Notes, Information Processing Society of Japan (IPSJ), http://id.ndl.go.jp/bib/6205223, Macroscopic properties of a bidirectional associative memory (BAM) are studied within a framework of S/N analysis called SCSNA. We obtained the relative capacity, which means the relative number of pattern pairs to be memorized and retrieved, as 0.199N, where N means the units in the system. We also derived dynamical properties by using the statistical neurodynamics and explained the property of BAM from transient process to equilibrium state consistently.
26 Jun. 2002
26 Jun. 2002- 26 Jun. 2002

Attempt to Classify Visualized Data Dependence
IWASAKA Asami; YAMAGUCHI Tomomi; SHOUNO Hayaru; SASAKURA Mariko; JOE Kazuki
Japanese, IPSJ SIG Notes, Information Processing Society of Japan (IPSJ), http://id.ndl.go.jp/bib/6134741, Recently, in the field of scientific calculation, the tendency to introduce parallel computer becomes increase. Therefore, converting methods from conventional sequential programs to correspondent parallel ones are required. However, vast amounts of skill and knowledge for the parallel programming are needed to programmer. In this study, we propose introducing the 3D visualization system called NaraView as those programming environment. Naraview analyze the data dependence in the programming code, and visualize their relations. Since the parallelizable code have typical features, NaraView would reveal them visually. We showed the effective parallelizing method, which have those typical features in the loop structure with concrete program code.
04 Mar. 2002
04 Mar. 2002- 04 Mar. 2002

The Design and Implementation of Unimodular Transformations for the Parallelizing Compiler PROMIS
ISHIUCHI Hisako; YAMAGUCHI Tomomi; SHOUNO Hayaru; JOE Kazuki
Japanese, IPSJ SIG Notes, Information Processing Society of Japan (IPSJ), http://id.ndl.go.jp/bib/6134746, For parallelizing compilers, many loop transformations have been proposed as optimization methods to exploit parallelism. However, since these methods have been designed separately, each method has its own conditions and effect to be applied. Therefore, even if we implement many loop transformations in a compiler, it is difficult to determine which combinatorial use of the transformations is optimal. Another transformation, which is called Unimodular, has the same effect to the combination of some transformations. Some transformation may get the optimal combination regarding to parallelism. In this paper, we describe the implementation of the unimodular transformation to the Parallelizing Compiler PROMIS, which is developed at the University of Illinois.
04 Mar. 2002
04 Mar. 2002- 04 Mar. 2002

Analysis of Telluric Current Data for Short - term Earthquake Prediction -Applications and Evaluations by ICA-
KOGANEYAMA Mika; SHOUNO Hayaru; NAGAO Toshiyasu; JOE Kazuki
Japanese, IPSJ SIG Notes, Information Processing Society of Japan (IPSJ), http://id.ndl.go.jp/bib/6009430, Seismic electric signals(SESs)are sometimes contained in telluric current data(TCD). The method of detecting SESs in TCD has attracted notice recently as an effective method for short-term earthquake prediction. However, since most of the TCD collected in Japan is affected by train noise, therefore detecting SESs in TCD itself is considered as an extremely arduous job. The goal of our research is to obtain a method for detecting SESs, which is difficult because of train noises. The SES and train noise are considered as independent source signal. Therefore we tried to apply ICA(Independent Component Analysis)to several TCDs. We have already confirmed that train noises could be separated from several TCDs at Matsushiro, Nagano. In this paper, we apply ICA to TCD at Sasadani, Fukui in which there are more train noises than Matsushiro, and evaluate the results.
19 Nov. 2001
19 Nov. 2001- 19 Nov. 2001

The Overview of the Parallelizing Compiler PROMIS - NWU
YAMAGUCHI Tomomi; ISHIUCHI Hisako; IWASAKA Asami; HANEDA Masayo; SHOUNO Hayaru; JOE Kazuki
Japanese, IPSJ SIG Notes, Information Processing Society of Japan (IPSJ), http://id.ndl.go.jp/bib/5866868, Compilers have been regarded as a cost expensive software. However, the implementation cost tends to be reduced recently by various new software development technologies. For example, software developments with object oriented technolgies require extremely inexpensive costs compared with conventional ones. Besides such software engineering approach, low-cost developments of compilers by the re-use of compiler components are expected. We are developing such a low-cost compiler by replacing the intermediate representation of PROMIS developed at UIUC with our new intermediate representation, aiming at other concepts for the compiler. In this report, we introduce the concept of the parallelizing compiler PROMIS-NWU which is under development at Nara Women's University.
25 Jul. 2001
25 Jul. 2001- 25 Jul. 2001

Naive mean field approximation to image restoration : Statistical-mechanical approach
SHOUNO Hayaru; WADA Koji; OKADA Masato
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, http://id.ndl.go.jp/bib/5871041, In the present paper, we treat image restoration in the framework of the Baysian inference. Recently, it is shown that under a certain criterion the MAP (Maximum A Posterior) estimate, which corresponds to minimization of energy, is outperformed by the MPM (Maximizer of the Posterior Marginals) estimate which is equivalent to a finite-temperature decoding method. Since a lot of computational time is needed for the MPM estimate to calculate the thermal averages, the mean field method, which is deterministic algorithm, is often utilized to avoid this difficulty. We give a statistical-mechanical analysis for the naive mean field approximation in the framework of image restoration. We compared the theoretical results with those of of computer simulation, and investigate the possibility of the naive mean field approximation.
20 Jul. 2001
20 Jul. 2001- 20 Jul. 2001

Detecting Seismic Electric Signals by LVQ Based Clustering
FUKUDA Kyoko; KOGANEYAMA Mika; SHOUNO Hayaru; NAGAO Toshiyasu; JOE Kazuki
Japanese, IPSJ SIG Notes, Information Processing Society of Japan (IPSJ), http://id.ndl.go.jp/bib/5834178, Aiming at short-term prediction of earthquakes, we have proposed the use of neural networks for analyzing telluric current data observed by the VAN method. We have already tried a telluric current data analysis method with Learning Vector Quantization. In this paper, we will show preliminary experimental results for categorization of telluric current data by its frequency for the Izu islands earthquakes in Japan.
26 Jun. 2001
26 Jun. 2001- 26 Jun. 2001

Analysis of BAM with SCSNA, and statistical neuro-dynamics
Shouno H.; Okada M.
Japanese, Meeting abstracts of the Physical Society of Japan, The Physical Society of Japan (JPS)
10 Sep. 2000
10 Sep. 2000- 10 Sep. 2000

27pU-7 Mean Field Theory of Image Restoration using Analog Network
SHOUNO H; OKADA M
Japanese, Meeting abstracts of the Physical Society of Japan, The Physical Society of Japan (JPS)
03 Sep. 1999
03 Sep. 1999- 03 Sep. 1999

Descrimination 3D object by Neocognitron
KOGA Kazuhisa; SHOUNO Hayaru; FUKUSHIMA Kunihiko; OKADA Masato
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, http://ci.nii.ac.jp/naid/110003233449, In macaque inferior temporal cortex, cells are known to respond to the complex feature, such as human face and so on. Logothetis et al reported most of cells in IT area have preferred view of the object after discrimination training, and the cell's responses decrease gradually by gradual deformations, such as rotation, magnification, and translation, from the preferred view. On the other hand, the visual form information is processed hierarchical by the ventral pathway, V1→V2→V4→IT. In this paper, we propose "neocognition" as a hierachical neural-network model of the pathway, and show the model can explain the Logothetis experiments. Moreover we discuss about the importance of the hierachy using the model.
21 Jan. 1999
21 Jan. 1999- 21 Jan. 1999

Neocognitoron with New Mechanism of Bend-detector
KIMURA Eiji; FUKUSHIMA Kunihiko; SHOUNO Hayaru
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, In visual pattern recognition, it is important to extract local features from input patterns. Linecrossings are one of such important features. We propose to use a neural network with a mechanism of disinhibition for the detection of line crosses, bends and end points. Bend-detector network which is uesd in the neocognitoron detects only curvature and end-points in inout pattern. We modified bend-detector to use disinhibition mechanism to detect not only end-points of lines and curvature also line-crossing feature which is important recogizing patterns. This modified neocognitoron has recognized handwriting degits in the ETL-1 data base with a recognition rate of about 98%.
20 Mar. 1998
20 Mar. 1998- 20 Mar. 1998

Selforganizing of C cell in Neocognitoron
YOSHIMOTO Kazuya; FUKUSHIMA Kunihiko; SHOUNO Hayaru
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, In the standard neocognitron, synaptic weights between feature extracting cells (S cells) and position invariant cells (C cell) are fixed and unmodifiable. This paper proposes a new learning rule by which the synaptic conection to Csell, as well as conection to Scells, are self-organized. Training paterns are moving line stimuli presented to the retina. In the conventional neocognitron, it was required to prepare in advance the architecture of `cell-plane'for each feature to be extracted. Through the new learning rule, however, the architecture of cell planes can be automatically self-organized in a network. The ability of the new learning rule is demonstrated by computer simulation.
19 Mar. 1998
19 Mar. 1998- 19 Mar. 1998

3D-Object Recognition using Neocognitron
SHOUNO Hayaru; OKADA Masato; FUKUSHIMA Kunihiko
Japanese, 日本神経回路学会全国大会講演論文集 = Annual conference of Japanese Neural Network Society, http://ci.nii.ac.jp/naid/10015797590
05 Nov. 1997
05 Nov. 1997- 05 Nov. 1997

Neural Network Model with A Mechanism of Disinhibition
KIMURA Eiji; FUKUSHIMA Kunihiko; SHOUNO Hayaru
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, In visual pattern recognition, it is important to extract local features from input patterns. Line-crossings are one of such important features. We propose to use a neural network with a mechanism of disinhibition for the detection of line crosses, bends and end points. Bend-detector network which is uesd in the neocognitoron detects only curvature and end-points in inout pattern. We modified bend-detector to use disinhibition mechanism to detect not only end-points of lines and curvature also line-crossing feature which is important recogizing patterns.
24 Jul. 1997
24 Jul. 1997- 24 Jul. 1997

Neocognitron applied to Handwritten Digit Recognition : Evaluation with Large Character Database
SHOUNO Hayaru; NAGAHARA Ken-ichi; FUKUSHIMA Kunihiko; OKADA Masato
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, http://ci.nii.ac.jp/naid/110003232688, The neocognitron is a hierachical neural network model of the visual processing system like mamals. It has ability to recognize pattern. In this paper, we improve the neocognitron for real world digit pattern recognition. The neocognitron classify a input digit pattern by the shape of the pattern. We introduce a new layer after the neocognitron system for pattern classifying the ten digit categories. The category classifying layer classify the output of the neocognitron to a category from '0' to '9'. We evaluate the performance of the neocognitron with category classifing network using ETL-1 database.
19 Jun. 1997
19 Jun. 1997- 19 Jun. 1997

Design and Implementation of the Neocognitron Class Library
OKAZAKI Tetsurou; SHOUNO Hayaru; FUKUSHIMA Kunihiko
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, The Neocognitron is a multi-layer neural-network model that has an ability to learn and recognize patterns. We have developed a set of class libraries, by which a simulator of the Neocognitron can easily be constructed. The purpose of this project is to provide a re-usable and extensible set of class libraries for the research on the Neocognitron. Documents and sources of these libraries implemented in C++ language are to be released to the public available. This article described the overview of the libraries and shows an example how a neocognitron simulator can be constructed with the libraries.
17 Mar. 1997
17 Mar. 1997- 17 Mar. 1997

Neocognitron Applied to Handwritten Digit Recognition : Evaluation with ETL Character Database
NAGAHARA Ken-ichi; SHOUNO Hayaru; FUKUSHIMA Kunihiko; OKADA Masato
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, http://ci.nii.ac.jp/naid/110003233139, The neocognitron is a neural network model which has the ability to recognize patterns. In our previous work, we obtained a recognition rate of 92.7% for handwritten digits in the ETL-1 database by using a high threshold for feature-extracting cells in the learning phase and a lower threshold in the recognition phase. In this paper, we changed learning method of the highest stage, and increased the number of the training patterns so that we obtained a recognition rate of 97.4%.
18 Mar. 1996
18 Mar. 1996- 18 Mar. 1996

選択的注意機構を用いた英文筆記体文字列認識
Hayaru Shouno; Kunihiko Fukushima
Oral presentation, Japanese, 日本神経回路学会,日本神経回路学会第6回全国大会講演論文集
Oct. 1995

Recognition Rate of the Neocognitoron for the ETL Character Database
NAGAHARA Ken-ichi; SHOUNO Hayaru; FUKUSHIMA Kunihiko
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, The neocognitoron is a neural network model which has the ability to recognize patterns. A previous work on the neocognitron showed that the recognition rate can be increased by using a high threshold for feature-extracting cells in the learning phase and a lower threshold in the recognition phase. The number of test sample of characters used in the previous work, however, was not large enough. In this paper, we use the ETL Character Database which contains a large numbers of handwritten numerals written by different persons. We confirmed that the use of different thresh-olds in learning and recognition phases is very effective for increasing the ability to recognize patterns robustly.
27 Jul. 1995
27 Jul. 1995- 27 Jul. 1995

Cursive Word Recognition Using Selective Attention
Shouno Hayaru; Fukushima Kunihiko
Japanese, IEICE technical report. Neurocomputing, The Institute of Electronics, Information and Communication Engineers, It is difficult to recognize connected characters in cursive handwriting,because the connecting stroke between characters varies depending to the combination of characters,thus producing slight variatins of the same character.The ′selective attention mo del′ proposed by Fukushima shows good results in recognizing compo site figures consisting of many patterns.Previously Imagawa et al. has attempted to use this model to recognize connected characters. On the other hand,it has been shown that the robustness of the ′ne ocognitron′ to deformed pattern is substantially in creased,when b end extractors and edge extractors are introduced.Therefore,we tried to introduce edge extractors and bend processors into the Imagawa′s system.Computer simulation has shown that this improved system has a higher ability to recognize connected characters.
25 Mar. 1994
25 Mar. 1994- 25 Mar. 1994

Algorithms I
Apr. 2017 - Present
The University of Electro-Communications

Media Science and Engineering Laboratory
Apr. 2015 - Present
The University of Electro-Communications

Media Computing System
Apr. 2015 - Present
The University of Electro-Communications

Programming Language Experiment
Apr. 2012 - Present
The University of Electro-Communications

Advanced Topics in Machine Learning
Apr. 2009 - Present
The University of Electro-Communications

Exercises in Algorithms and Data Structures
Apr. 2011 - Mar. 2015
The University of Electro-Communications

アルゴリズム基礎論演習
Apr. 2008 - Mar. 2011
電気通信大学

Basic Algorithms
Apr. 2008 - Mar. 2011
The University of Electro-Communications

データ駆動科学F
熊本大学

統計的なパターン認識および機械学習
熊本大学

機械学習・深層学習論
東京学芸大学

生物工学特論D: 視覚情報処理とニューラルネットワークモデル
大阪大学

情報科学特別講義: ニューラルネットを用いた画像処理
山口大学

基礎プログラミングおよび演習
電気通信大学

総合情報学基礎
電気通信大学

電子情報通信学会

日本神経回路学会

日本物理学会

情報処理学会

IEEE

ディープラーニングのホワイトボックス化に関する研究
Masato Okada
01 Apr. 2018 - 31 Mar. 2023

スペクトル分解のためのFPGAを用いた確率的サンプリングマシンの開発
01 Apr. 2019 - 31 Mar. 2022

視覚心理に基づくテクスチャ特徴表現と深層特徴表現のマッピング
庄野 逸
Principal investigator
01 Apr. 2019 - 31 Mar. 2021

スパース基底表現を用いた断層画像再構成アルゴリズムの開発
庄野 逸
Principal investigator
01 Apr. 2016 - 31 Mar. 2019

多元計算解剖学における形態情報統合の基盤技術
Akinobu Shimizu
01 Apr. 2016 - 31 Mar. 2018

Deep Learning を用いたスパーステクスチャ画像解析手法の確立
庄野 逸
Principal investigator
01 Apr. 2016 - 31 Mar. 2018

相関スパース表現 Deep Architecture によるテクスチャ解析
01 Apr. 2014 - 31 Mar. 2016

Bayesian medical image integration with statistical mechanical approach
01 Apr. 2013 - 31 Mar. 2016

計算解剖モデルの支援診断とオートプシー・イメージング支援応用
01 Apr. 2009 - 31 Mar. 2014

3 Dimensional Image Reconstruction Algorithm using Statstical Mechanics of Information
01 Apr. 2009 - 31 Mar. 2012

Machine learning via fusion of discriminative and mean field models and its application to image recognition
TAKAHASHI Haruhisa; HOTTA Kazuhiro; SHOUNO Hayaru; RAMESWER Debnathd
Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research, The University of Electro-Communications, Grant-in-Aid for Scientific Research (C), In machine learning, we have two typical models, i. e., generative and discriminative models. We aimed to combine both of these models in order to obtain more elaborated machine learning models. Two main results are obtained as follows. Firstly, we designed a special discriminative random field and applied it to high performance video image classification. Secondly, we modeled a kernel random field which is constructed as random field with kernels, and successively applied to scene classification., 21500213
2009 - 2011

A study of application for medical imaging algorithm based on probabilistic information processing
SHOUNO Hayaru
Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research, Grant-in-Aid for Young Scientists (B), 新たな医療画像アプリケーションを作成するための理論的な枠組みの構築と,医療診断を支援するようなプロトタイプ作成を主眼に置き研究を行った.特に近年発展してきているソフトコンピューティング分野の手法を援用することにより,観測ノイズによる画像の不鮮明性を考慮にいれた上での画像の特性を調査し,より効果的なアプリケーションの開発を行うことが出来たと考えられる., 18700223
2006 - 2008

Development of Computeraided Diagnosis System for Evaluation of Regional Pulmonary Blood Flow by use of Different Thee Phase of MDCT Images
KIDO Shoji; SHOUNO Hayaru
Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research, Yamaguchi University, Grant-in-Aid for Scientific Research (B), The outline of our research results is summarized to the following four items : 1. Regional Pulmonary Blood Flow Analysis We have developed a subtraction algorithm for evaluation of pulmonary blood flow abnormalities based on non-solid registration algorithm by use of enhanced and non-enhanced MDCT images. We have developed an extraction algorithm of peripheral pulmonary vessels from volume MDCT data. Our algorithm is useful for diagnosis of diffuse lung diseases. We have also developed a detection algorithm of acute pulmonary embolisms. 2. Function Analysis of Pulmonary Respiration COPD is one of targets for function analysis of pulmonary respiration. For the diagnosis of COPD, we have developed a radial measurement of bronchus. And also, for the evaluation of emphysema, we have developed an algorithm of lobar segmentation. Moreover, we have developed a measurement algorithm of displacements for ribs and diaphragms by use of inspiratory and expiratioy MDDCT images. 3. Analysis of Diffuse Lung Disease Diffuse lung disease is one of targets for our study. We have developed an algorithm of diffuse lung analysis for whole lungs obtained from MDCT volume data. In our algorithm, whole lungs are diagnosed as one of seven diffuse lung disease patterns. 4. Analysis of Auscultation Sounds For evaluation of pulmonary diseases, such as pneumonia and COPD, we have developed an algorithm of auscultation sounds. We have achieved more than 90% recognition rate for differentiation of auscultation sounds., 17300174
2005 - 2007

Intelligent CAD based on understanding of local pathological structures
KIDO Shoji; SHOUNO Hayaru; MATSUMOTO Tsuneo; MATSUNAGA Naofumi
Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research, Yamaguchi University, Grant-in-Aid for Scientific Research on Priority Areas, The outline of our research results is summarized to the following four items: 1. Pattern classification of diffuse lung diseases using MDCT volume data Diffuse lung disease is one of targets for our study. We have developed an algorithm of diffuse lung analysis for whole lungs obtained from MDCT volume data. In our algorithm, whole lungs are diagnosed as one of seven diffuse lung disease patterns. 2. Function Analysis of Pulmonary Respiration COPD is one of targets for function analysis of pulmonary respiration. For the diagnosis of COPD, we have developed a radial measurement of bronchus. And also, for the evaluation of emphysema, we have developed an algorithm of lobar segmentation. Moreover, we have developed a measurement algorithm of displacements for ribs and diaphragms by use of inspiratory and expiratioy MDDCT images. 3. Development of a CAD platform We have developed a CAD platform named "MARIMO" for supporting image diagnoses of radiologists for multi-organ and multi-diseases. 4. Analysis of Auscultation Sounds For evaluation of diffuse lung diseases, such as pneumonia and COPD, we have developed an algorithm of auscultation sounds. We have achieved more than 90% recognition rate for differentiation of auscultation sounds. And also, we compared the auscultation sound with MDCT images., 15070208
2003 - 2006

統計力学に基づいたデジタル情報処理へのアナログ的アプローチに関する研究
庄野 逸
日本学術振興会, 科学研究費助成事業, 山口大学, 若手研究(B), 画像・信号修復の統計力学的な解析に関する研究、および、神経信号の計測時に用いられる多重電極センサから得られる信号解析方法の統計力学的な評価を行った.今年度も昨年度に引き続き,多重電極による信号解析に関する研究において対数線形モデルを用いたスパイク状信号の解析手法を中心に研究を行った.研究のメインとなる軸はベイズ推定に基づく信号の統計力学的な議論に有り,様々な方々と有意義な議論を行った上で研究を遂行した.対数線形モデルは質的情報の多変量解析を行なう場合に用いられるモデルであり,様々な分野において応用が試みられている.一方,神経科学の分野においては,多重電極による神経細胞群活動の同時計測が実用化の段階にまで至っている.今後の計測技術を見越すと同時計測技術は,数百個〜数千個単位のオーダーの同時計測が可能になると思われる.これはニューロン数が非常に多い場合の解析手法を確立する必要があることを意味し,スパイク信号解析に用いられる対数線形モデルを,ニューロン数が無限大の系で扱うことは今後重要な課題になると考えられる.本研究では対数線形モデルによって記述される系を統計物理で用いられるようなイジングスピンからなる系とみなし,ニューロン数が無限におけるハイパーパラメータ推定の統計力学的解析を行なった.解析を行なうに際し,スパイク間には一様な高次相互作用を仮定し,この高次相互作用の係数をハイパーパラメータとして推定を行なった.私は観測過程において信号が反転するようなノイズが混入されるモデルを用い,相互作用のハイパーパラメータを推定する際には山登り法とEMアルゴリズムを適用した場合の解析を行なった., 15700192
2003 - 2005

Dynamic Processing of Visual Patterns
FUKUSHIMA Kunihiko
Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research, Grant-in-Aid for Scientific Research (A)., Aiming to develop new design principles for visual information processing systems of the next generation, we have concentrated our research on the active and dynamic processes in the visual system of the biological brain. We used modeling approach to uncover the mechanism of the brain and to design advanced systems for visual pattern recognition. We have performed various researches in parallel and have obtained the following results. 1. Neural network model that can recognize faces from complex backgraound. It can focus attention to and segment facial components (eyes and mouth) from the recognized face. 2. Neural network model that can recognize partly occluded patterns correctly. 3. Neocognitron of a new version for recognizing handwritten digits in the real world. The neocognitron, which we have developed previously, is a pattern recognition system whose architecture has been suggested from the mammalian visual system. The recognition rate, which varies depending on the size of training set, was over 98% when we used 3000 characters for the training. 4. New learning rule by which cells with shift-invariant receptive fields are self-organized. With this learning rule, cells similar to simple and complex cells in the primary visual cortex are generated in a network. 5. Neural network model that can memorize and recall spatial maps. The model emulates a situation where a person memorizes and recalls spatial maps when he moves around in a two-dimensional space. The model memorizes fragmentary maps, but can retrieve an image covering a wide area seamlessly by a continuous chain process of recalling. 6. Stereo algorithm that extracts a depth cue from interocularly unpaired points., 09308010
1997 - 2000

多層構造を持つ神経回路モデルによる物体認識
庄野 逸
日本学術振興会, 科学研究費助成事業, 大阪大学, 奨励研究(A), 我々は4層の多層構造を持つ神経回路モデルを構築し、この神経回路モデルに対してコンピュータグラフィクスなどを用いて3D図形を生成してシミュレーションによる認識実験を行なった。神経回路モデルの詳細部分は実際の脳の神経系の反応特性と比較し、最終層での反応特性がある程度一致するまでパラメータチューニングを行なった。過去の文献よると、高次視覚領野(サルのIT野)などに複雑な図形を与えた場合の反応では、多数の細胞が、物体のある一部分から見た特定の像に依存した反応特性を示すことが知られている。さらに物体を回転させるなどして、像を変形させると、徐々に反応のレベルが落ちてくることが知られていることがわかった。我々は、この神経回路モデルに3次元物体を学習させ、自己組織化させた上で、学習した像から、物体を回転・拡大・縮小・平行移動などの変形を行い、どのような特性を示すのかを検討した。我々の構築した神経回路モデルでは生理実験で調べられている限りの上での不整合は見られなかった。その結果を確認した上で我々はどのような空間フィルタが形成されているかを調べ、物体の角や、T-ジャンクションと呼ばれる部分のフィルタが多数形成されているのを確認した。これらのフィルタは2次元物体を識別する際に重要な特徴と考えられており、3次元物体の認識を行う際にも重要な特徴であるということが確認できた。, 10780231
1998 - 1999

多層構造をもつ.神経回路モデルにおけるトポグラフィックマッピングの形成
庄野 逸
日本学術振興会, 科学研究費助成事業, 大阪大学, 奨励研究(A), 哺乳類等の大脳皮質、特に感覚情報処理の初期段階を受け持つ感覚野と呼ばれる領域と、筋肉に指令を発する運動野と呼ばれる領域等においては連続的な機能地図(トポグラフィックマッピング)が存在する。簡単にいえば、感覚野においては近くにある神経細胞は似たような刺激に反応し、運動野においては近くの細胞は近くの筋肉に指令を送っているということである。特に最も良く研究されているのは感覚野である初期視覚領野の機能地図形に関してである。オプチカルレコーディング等の計測技術の進歩により皮質地図,特に哺乳類の視覚1次野の特徴が解明されてきている。哺乳類の大脳皮質の初期視覚野においては,様々な方位を持つ線分に対する反応する機能地図(マップ)が発見されている.近年では,方位だけでなく様々な方向に線分を動かした時に反応する機能マップがferretや猫の17野等で発見されており,方向に関するマップと方位に関するマップとの関連が議論されている.ここで述べる“方位(orientation)"とは,提示された線分の傾き(0〜π)を表し,“方向(direction)"とは,その線分の移動方向まで区別した言いかたである.したがって一つの方位には正反対の二つの方向が対応している.皮質上に形成される方位マップと方向マップとは互いに矛盾のないようにできているはずであるが,これらのマップの関係には二通りのありようが可能である.一つは方位マップ上で,ある方位に対応する一個の領域が一つの方向だけに反応し,それと正反対の方向に反応する細胞は,別の場所に存在しているという可能性である.もう一つの可能性は,一つの方位選択領域が正反対の向きを持つ2つの方向領域に分割されているという可能性である.すなわち,正反対の方向に反応する領域が隣接して一つの方位領域をなしている場合である.Feretや猫17野で観測されているのは後者の機能地図である。このような機能地図を形成する自己組織化モデルとしてKohonenのSOMが挙げられる.本研究ではこのSOMを用いて方向及び方位を含めたマップを2次元平面上に形成し,ferretや猫の17野で発見されているような機能地図を形成することに成功した., 08780358
1996 - 1996

Research on Visual Pattern Recognition with Hierarchical Neural Networks
FUKUSHIMA Kunihiko; OKADA Masato; SHOUNO Hayaru
Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research, Osaka University, Grant-in-Aid for Scientific Research (A), Aiming to develop new design principles for visual information processing systems of the next generation, we have concentrated our research on the active processes in the visual system of the biological brain. We used modeling approach to solve the mechanism of the brain, and proposed neural network models explaining various functions related to active vision. We also tried to design visual pattern recongnition systems using the results of the modeling research. We have performed various researches in parallel and have obtained the following results. (1) Neural network model that has two separate channels processing form and motion information. The model can solve the binding problem by the function of selective attention. (2) Neural network model of binocular cells. The model includes far-cells, near-cells, and fine-tuned cells. We also proposed a theory that can estimate the depth of an object occluded from one eye, and have shown that the results obtained from our theory coincide with the psychological experiments. (3) Eye movement model with non-uniform receptive fields. (4) Neural network model of spatial memory. The model memorizes the fragmentary maps of external world, and can recall a map of a wide area by a chain process of recalling. (5) Training neocognitron to recognize handwritten characters in the real world. The neocognitron, which we have developed previously, is a pattern recognition system whose architecture has been suggested from the mammalian visual system. We trained the neocognitron using a large-scale data base of handwritten digits (ETL-1), and obtained a recognition rate higher than 98%. (6) Theoretical analysis of the correlation matrix memory., 07408005
1995 - 1996

学習装置、学習方法及び学習プログラム
Patent right, 鈴木 聡志, 増村 亮, 澤田 雅人, 安藤 厚志, 牧島 直輝, 庄野 逸, 特願2022-114303, Date applied: 15 Jul. 2022

学習装置、学習方法及び学習プログラム
Patent right, 鈴木 聡志, 増村 亮, 澤田 雅人, 安藤 厚志, 牧島 直輝, 庄野 逸, 特願2022-114304, Date applied: 15 Jul. 2022

モデル評価装置、モデル評価方法およびモデル評価プログラム
Patent right, 鈴木 聡志, 樋口 陽光, 庄野 逸, 特願2021-158633, Date applied: 29 Sep. 2021, Nippon Telegraph and Telephone Corporation

情報処理装置、プログラム、及び情報処理装置の動作方法
Patent right, 高杉順子, 中村理恵, 黒沢正治, 庄野逸, 特願2020-197515, Date applied: 27 Nov. 2020, 株式会社コーセー , 国立大学法人電気通信大学

画像処理方法、画像処理装置及びプログラム
Patent right, 庄野逸, 鈴木聡志, 谷田隆一, 木全英明, 特願2020-088398, Date applied: 20 May 2020, Nippon Telegraph and Telephone Corporation

放送大学 データサイエンスの技術
Appearance, 放送大学, ニューラルネットワーク概論
01 Oct. 2021