Paper Abstract and Keywords |
Presentation |
2020-09-03 11:00
[Short Paper]
Quantitative analysis of epicardial adipose tissue by two-stage segmentation network and its system development Takayuiki Nagata, Yutaro Iwamoto, Zhao Ziyu (Ritsumeikan Univ), Yuji Tezuka, Hiroki Okada, Kiyosumi Maeda, Atsuyuki Wada, Atsunori Kashiwagi (Kusatsu General Hospital), Yen-Wei Chen (Ritsumeikan Univ) MI2020-23 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
Diabetes is thought to lead to vascular disease and arteriosclerosis, and there is a need for early detection and treatment. Recent studies have shown that epicardial adipose tissue (EAT), which is inside the epicardium surrounding the heart, is strongly associated with diabetes. Due to the difficulty of accurate detection by the network of epicardium used for isolation, in this paper, we propose a two-stage segmentation network that allows accurate prediction of regions within the epicardium and EAT detection using a hostile generation network, even if the detection of epicardium is not sufficient. We also developed a system that can predict, correct, and re-predict EAT simply by inputting a CT image of the heart region. In the proposed method, the median value of the Dice coefficient was 88.5% in 14 cases and the average was 86.6%, which exceeded the median value of 82.3% in 250 cases in the previous study. In addition, when the developed system was used, it was possible to detect in about 1 minute per case and display it. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
Diabetes / Epicardial adipose tissue / Segmentation / CT image / Deep learning / Adversarial generation network / Developed system / |
Reference Info. |
IEICE Tech. Rep., vol. 120, no. 156, MI2020-23, pp. 27-30, Sept. 2020. |
Paper # |
MI2020-23 |
Date of Issue |
2020-08-27 (MI) |
ISSN |
Online edition: ISSN 2432-6380 |
Copyright and reproduction |
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
Download PDF |
MI2020-23 |
|