(주) 뷰노

(주) 뷰노

문의사항 남기기
KR EN
VUNO MED INC.
Publications
BACK TO LIST
MICCAI 2022 Workshop on Medical Image Learning with Limited and Noisy Data

Abstraction in Pixel-wise Noisy Annotations Can Guide Attention to Improve Prostate Cancer Grade Assessment

  • Sep. 2022

Abstract:

Assessing prostate cancer grade from whole slide images (WSIs) is a challenging task. While both slide-wise and pixel-wise annotations are available, the latter suffers from noise. Multiple instance learning (MIL) is a widely used method to train deep neural networks using WSI annotations. In this work, we propose a method to enhance MIL performance by deriving weak supervisory signals from pixel-wise annotations to effectively reduce noise while maintaining fine-grained information. This auxiliary signal can be derived in various levels of hierarchy, all of which have been investigated. Comparisons with strong MIL baselines on the PANDA dataset demonstrate the effectiveness of each component to complement MIL performance. For 2,097 test WSIs, accuracy (Acc), the quadratic weighted kappa score (QWK), and Spearman coefficient were increased by 0.71%, 5.77%, and 6.06%, respectively, while the mean absolute error (MAE) was decreased by 14.83%. We believe that the method has great potential for appropriate usage of noisy pixel-wise annotations.

Link

https://link.springer.com/chapter/10.1007/978-3-031-16760-7_3

Author

-

Tag in

#pathology

이전글
Volume is All You Need: Improving Multi-task Multiple Instance Learning for WMH Segmentation and Severity Estimation 2022. 10. 06
다음글
Development and Validation of a Deep Learning-Based Automatic Classification Algorithm for the Medial Temporal Lobe Atrophy Score using a Multi-Modality Cascade Transformer 2025. 06. 25
List

이 웹사이트는 원활한 서비스 제공을 위하여 당사 개인정보 처리방침에 명시된 쿠키를 사용합니다.

만일 당사의 쿠키 사용 정책에 대해 더 알고 싶으신 경우에는 개인정보 처리방침 을 참고하여 주시기 바랍니다.