Paper Abstract and Keywords |
Presentation |
2024-03-03 16:42
[Short Paper]
Using Label Uncertainty for Learning Cell Nuclei Type Classifier with Strongly Noisy Supervised Signals Shingo Koide, Mauricio Kugler, Tatsuya Yokota (NIT), Koichi Ohshima, Hiroaki Miyoshi, Miharu Nagaishi (Kurume Univ.), Noriaki Hashimoto (RIKEN), Ichiro Takeuchi (Nagoya Univ.), Hidekata Hontani (NIT) MI2023-57 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
In this study, we construct a type classifier for cell nuclei of malignant lymphomas. Labelling by type is not easy, even for pathologists, and labels can contain a large number of errors.We observed pathologists labelling cell nuclei by type and found that they were often unable to decide between the two types.Therefore, we asked pathologists to record not only the final decision result when they were unsure about labelling, but also the names of the types that they discarded after hesitation.The proposed method approximates the generation probability of mislabeling by using the teacher's signal including the rejected labels. Then, the cell nucleus type discriminator is trained considering this probability of mislabel generation. This paper reports the results of a comparison experiment with a method that does not consider the generation probability of mislabels, confirming the effectiveness of the proposed method. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
Image classification / Label Noise / Malignant lymphoma / / / / / |
Reference Info. |
IEICE Tech. Rep., vol. 123, no. 411, MI2023-57, pp. 79-80, March 2024. |
Paper # |
MI2023-57 |
Date of Issue |
2024-02-25 (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 |
MI2023-57 |
|