Faculty Profiles - ODA, Masahiro

Presentations -

Division display 141 - 160 of about 285 ／ All the affair displays >>

Visualising decision-reasoning regions in computer-aided pathological pattern diagnosis of endoscytoscopic images based on CNN weights analysis International conference

Hayato Itoh, Zhongyang Lu, Yuichi Mori, Masashi Misawa, Masahiro Oda, Shin-ei Kudo, Kensaku Mori

SPIE Medical Imaging 2020 2020.2.17

Visualizing intestines for diagnostic assistance of ileus based on intestinal region segmentation from 3D CT images International conference

Hirohisa Oda, Kohei Nishio, Takayuki Kitasaka, Hizuru Amano, Aitaro Takimoto, Akinari Hinoki, Hiroo Uchida, Kojiro Suzuki, Hayato Itoh, Masahiro Oda, Kensaku Mori

SPIE Medical Imaging 2020 2020.2.17

Multi-modality super-resolution loss for GAN-based super-resolution of clinical CT images using micro CT image database International conference

Tong Zheng, Hirohisa Oda, Takayasu Moriya, Takaaki Sugino, Shota Nakamura, Masahiro Oda, Masaki Mori, Hirotsugu Takabatake, Hiroshi Natori, Kensaku Mori

SPIE Medical Imaging 2020 2020.2.17

深層学習時代の医用画像処理におけるデータと計算環境 Invited

小田昌宏

PCCC19「来たるべきSociety 5.0時代に向けて」（第19回PCクラスタシンポジウム） 2019.12.13

Generative Adversarial Networks Showcase: Their Mechanisms and Radiological Applications International conference

Masahiro Oda, Hirohisa Oda, Kanako K. Kumamaru, Shigeki Aoki, Hiroshi Natori, Kensaku Mori, Masaki Mori, Hirotsugu Takabatake

Radiological Society of North America (RSNA) 2019

Realistic virtual endoscopic image generation method using image domain translation

Masahiro Oda,Kiyohito Tanaka,Hirotsugu Takabatake,Masaki Mori, Hiroshi Natori,Kensaku Mori

The 28th Annual Congress of Japan Society of Computer Aided Surgery

Event date： 2019.11

Language：Japanese Presentation type：Oral presentation (general)

Venue：University of Tokyo Country：Japan

We propose a realistic virtual endoscopic image generation method for endoscope simulators. Endoscope simulators are used to train endoscopists. However, less realistic visualization of simulators reduces effect of training. Main differences of virtual endoscopic images generated by simulators and real endoscopic images are textures and light reflections on the surface of organs. The differences should be reduced to improve reality of virtual images. This paper proposes a method to improve reality of virtual endoscopic images using a virtual to real endoscopic image domain translation technique. In the image domain translation, the structure of fully convolutional network (FCN) that performs image translation has relation to the reality of translated images. We experimentally confirmed reality of translated images using three different FCNs. From the result, use of larger numbers of kernels and layers resulted in generation of realistic images.

深層学習を用いた医用画像処理研究の最前線 Invited

小田昌宏

第47回日本放射線技術学会秋季学術大会（第86回画像部会教育講演）

JAMITハンズオンセミナー

原武史, 中田典生, 小田昌宏, 福岡大輔, 田中理恵, 中山良平

第38回日本医用画像工学会大会

Estimation of Contrasted Abdominal CT Volume from Non-contrasted CT Volume using Generative Adversarial Frameworks

Masahiro Oda, Kanako Kumamaru, Shigeki Aoki, Kensaku Mori

The 38th JAMIT Annual Meeting

Event date： 2019.7

Language：Japanese Presentation type：Poster presentation

Venue：Nara Kasugano International Forum Country：Japan

We propose an estimation method of an abdominal contrasted CT volume from a non-contrasted CT volume. In diagnosis and treatment, contrasted CT volumes are taken to confirm anatomical structure and abnormal regions related to the blood vessels. However, administration of contrast agents causes dyspnea and cardiac arrest in some cases. In this paper, we propose an estimation method of contrasted images from non-contrasted images in abdominal CT volumes. As the estimation method, we employ an image regression model using a fully convolutional network (FCN). We trained the FCN in a direct-training and indirect-trainings using generative adversarial frameworks (pix2pix, CycleGAN). In our experiments using abdominal CT volumes, contrasted CT-like volumes having good-quality were generated when we used the CycleGAN.

Non-contrast to contrasted abdominal CT volume regression using fully convolutional network International conference

Masahiro Oda, Kanako K. Kumamaru, Shigeki Aoki, Kensaku Mori

Computer Assisted Radiology and Surgery (CARS) 2019

AI and Machine Learning in Medical Image Engineering Invited

Masahiro Oda

The 63rd Annual Conference of the Institute of Systems, Control and Information Engineers (SCI'19)

Event date： 2019.5

Language：Japanese Presentation type：Oral presentation (invited, special)

Venue：Central Electric Club Country：Japan

In the medical image processing field, utilization of AI and machine learning techniques are getting popular. Among AI and machine learning techniques, the deep learning significantly contributes to improve performances of medical image processing methods. This paper introduces some automated segmentation and lesion detection techniques including conventional and deep learning-based methods. Improvements of their performances obtained by using deep learning techniques are also presented. Additionally, two problems that medical image processing researches who use AI and machine learning techniques facing are introduced.

Development of AI assisted diagnosis and treatment powered by integrated use of various data Invited

Masahiro Oda

The 92th Annual Meeting of the Japan Orthopaedic Association

AIプログラミング基礎編，AIプログラミング実践編 Invited

小田昌宏

えひめAI・IoT研究会技術セミナー

AIを活用した画像処理について Invited

小田昌宏

えひめAI・IoT研究会技術セミナー

Colonoscope tracking method based on shape estimation network International conference

Masahiro Oda, Holger R. Roth, Takayuki Kitasaka, Kazuhiro Furukawa, Yoshiki Hirooka, Nassir Navab, Kensaku Mori

SPIE Medical Imaging 2019

Event date： 2019.2

Language：English Presentation type：Oral presentation (general)

Venue：San Diego, California, USA Country：United States

This paper presents a colonoscope tracking method utilizing a colon shape estimation method. CT colonography is used as a less-invasive colon diagnosis method. If colonic polyps or early-stage cancers are found, they are removed in a colonoscopic examination. In the colonoscopic examination, understanding where the colonoscope running in the colon is difficult. A colonoscope navigation system is necessary to reduce overlooking of polyps. We propose a colonoscope tracking method for navigation systems. Previous colonoscope tracking methods caused large tracking errors because they do not consider deformations of the colon during colonoscope insertions. We utilize the shape estimation network (SEN), which estimates deformed colon shape during colonoscope insertions. The SEN is a neural network containing long short-term memory (LSTM) layer. To perform colon shape estimation suitable to the real clinical situation, we trained the SEN using data obtained during colonoscope operations of physicians. The proposed tracking method performs mapping of the colonoscope tip position to a position in the colon using estimation results of the SEN. We evaluated the proposed method in a phantom study. We conrmed that tracking errors of the proposed method was enough small to perform navigation in the ascending, transverse, and descending colons.

DOI： 10.1117/12.2512729

Report on MICCAI 2018

Masahiro Oda, Yoshito Otake, Hayato Itoh, Takaaki Sugino, Atsushi Saito, Ryo Furukawa, Takashi Onishi, Atsushi Imiya, Kensaku Mori

Deep Learning Sample Code Collection for Medical Image Processing: DMED

Masahiro Oda, Takeshi Hara, Kensaku Mori

The 27th Annual Congress of Japan Society of Computer Aided Surgery

Event date： 2018.11

Language：Japanese Presentation type：Poster presentation

Country：Japan

Deep learning techniques are widely used in both of computer aided surgery (CAS) and diagnosis (CAD) fields. Many software to use deep learning are available. However, in the medical assisting research field, we need to implement new deep learning-based methods. It is difficult for researchers who are not specialized in image processing or machine learning. To accelerate use of deep learning for all researchers, we developed the deep learning sample code collection for medical image processing (DMED). DMED enables easy use of key deep learning-based methods including medical image processing. Current DMED includes sample codes for image classification, segmentation, and superresolution. Different from the previous deep learning library such as the NiftyNet and the DLTK, the DMED performs training processes in reasonable processing time without GPU. We expect the DMED accelerates use of deep learning-based image processing methods in the CAS and CAD fields.

Deep learning for medical images in surgical field

Masahiro Oda

The 27th Annual Congress of Japan Society of Computer Aided Surgery

Event date： 2018.11

Language：Japanese Presentation type：Oral presentation (invited, special)

Country：Japan

Deep learning contributes growing of the information processing field. It is used to achieve computer systems having high recognition performances. Traditionally, recognition algorithm was individually developed for each task. In contrast to them, a common model or a framework is applicable to many recognition tasks in the deep learning field. This property contributes acceleration of research and development. In this session, we explain how to perform image processing using deep learning in hands-on style. How to develop and execute basic deep learning-based image processing methods including image classification and segmentation are explained. Programs of these methods will run on attendee's computers. Attendee are requested to install software and download data to their own computers prior to joining the hands-on.

AIによるコンピュータ外科の始め方 - 研究環境構築法 Invited

小田昌宏

第27回日本コンピュータ外科学会大会

Colon Shape Estimation Method for Colonoscope Tracking using Recurrent Neural Networks International conference

Masahiro Oda, Holger Roth, Takayuki Kitasaka, Kazuhiro Furukawa, Ryoji Miyahara, Yoshiki Hirooka, Hidemi Goto, Nassir Navab, Kensaku Mori

MICCAI 2018

　 PREV - NEXT 　

<ODA, Masahiro>

Presentations -