Paper Abstract and Keywords |
Presentation |
2021-03-16 13:45
Deep Learning prediction of lung transplant rejection from FDG-PET and visualization of the basis for the decision Keisuke Hori (Chiba Univ.), Yuma Iwao, Miwako Takahashi (QST), Haruhiko Shiiya (UTokyo/Hokkaido Univ.), Masaaki Sato (UTokyo), Taiga Yamaya (QST) MI2020-73 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
We conducted a study to correlate FDG-PET images with pathological diagnosis of inflammation in lung transplantation (LTx) model rats, with the task of detecting early signs of chronic rejection from FDG-PET after LTx. In this study, we used VGG16 pre-trained on ImageNet to predict the pathological diagnosis at 6 weeks from FDG-PET images at 3 weeks after LTx, and the most accurate epoch was 96% in sensitivity and 91% in specificity. Furthermore, we analyzed the basis for deep learning from the change in prediction accuracy by reducing the features in the input image, and showed that the information in the lower part of the left lung may be important for pathological prediction. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
Lung Transplantation / FDG-PET / Deep Learning / / / / / |
Reference Info. |
IEICE Tech. Rep., vol. 120, no. 431, MI2020-73, pp. 108-111, March 2021. |
Paper # |
MI2020-73 |
Date of Issue |
2021-03-08 (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 |
MI2020-73 |
|