Paper Abstract and Keywords |
Presentation |
2021-07-09 11:00
[Short Paper]
Construction of Subtype Classifier for Malignant Lymphoma based on H&E-stained Images using Immuno-stainning Data Yuki Hirono (NIT), Noriaki Hashimoto (RIKEN), Kugler Mauricio, Tatsuya Yokota (NIT), Miharu Nagaishi (Kurume Univ.), Hiroaki Miyoshi, Koichi Oshima (Kurume Univ./JSP), Ichiro Takeuchi (NIT/RIKEN), Hidekata Hontani (NIT) MI2021-16 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
In pathological diagnosis of malignant lymphoma, a HE image is observed at first and then a set of immunostained images are observed to determine the subtype. Observing the HE image, pathologists infer the candidates of the subtypes, determine the set of immunostains needed for identifying the subtype, and finally identify the subtype by observing if the specimen is positively stained by each of the immunostains. The information from HE images is the candidates of the subtypes and a set of immunostains needed for the subtype identification. The proposed method hence constructs a decision tree for inferring the set of subtype candidates and the set of the immunostains from an input HE image. Each node of the decision tree infers if a set of specific immunostains is needed for the subtype identification. We used a set of pairs of a HE image and the text data that describes the diagnosed subtype and the set of immunostains. The multiple-instance learning (MIL) is employed for the training as we have no labels indicating the cancerous regions in the HE images. The outline of the proposed method and some results of initial studies are reported. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
pathological image / malignant lymphoma / H&E-stained image / immuno-staining data / neural network / multiple instance learning / / |
Reference Info. |
IEICE Tech. Rep., vol. 121, no. 98, MI2021-16, pp. 31-32, July 2021. |
Paper # |
MI2021-16 |
Date of Issue |
2021-07-01 (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 |
MI2021-16 |
|