Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Denoising Mutual Knowledge Distillation in Bi-Directional Multiple Instance Learning

Chen Shu, Boyu Fu, Yiman Li, Ting Yin, Wenchuan Zhang, Jie Chen, Yuhao Yi, Hong Bu

TL;DR

This work tackles Whole Slide Image classification with imperfect, weak supervision by introducing a bi-directional dual-branch MIL framework that mutualizes bag- and instance-level learning. Each branch provides soft pseudo-labels to supervise the other, while self-confidence losses and a carefully balanced loss design mitigate noise from pseudo-labels. A scheduled training regime keeps bag- and instance-branch optimization in sync, enabling robust improvements in both bag- and instance-level predictions, demonstrated on CAMELYON16 and TCGA-NSCLC. The approach advances MIL robustness in digital pathology by leveraging weak-to-strong generalization to denoise supervision signals and refine discriminative patch representations.

Abstract

Multiple Instance Learning is the predominant method for Whole Slide Image classification in digital pathology, enabling the use of slide-level labels to supervise model training. Although MIL eliminates the tedious fine-grained annotation process for supervised learning, whether it can learn accurate bag- and instance-level classifiers remains a question. To address the issue, instance-level classifiers and instance masks were incorporated to ground the prediction on supporting patches. These methods, while practically improving the performance of MIL methods, may potentially introduce noisy labels. We propose to bridge the gap between commonly used MIL and fully supervised learning by augmenting both the bag- and instance-level learning processes with pseudo-label correction capabilities elicited from weak to strong generalization techniques. The proposed algorithm improves the performance of dual-level MIL algorithms on both bag- and instance-level predictions. Experiments on public pathology datasets showcase the advantage of the proposed methods.

Denoising Mutual Knowledge Distillation in Bi-Directional Multiple Instance Learning

TL;DR

This work tackles Whole Slide Image classification with imperfect, weak supervision by introducing a bi-directional dual-branch MIL framework that mutualizes bag- and instance-level learning. Each branch provides soft pseudo-labels to supervise the other, while self-confidence losses and a carefully balanced loss design mitigate noise from pseudo-labels. A scheduled training regime keeps bag- and instance-branch optimization in sync, enabling robust improvements in both bag- and instance-level predictions, demonstrated on CAMELYON16 and TCGA-NSCLC. The approach advances MIL robustness in digital pathology by leveraging weak-to-strong generalization to denoise supervision signals and refine discriminative patch representations.

Abstract

Multiple Instance Learning is the predominant method for Whole Slide Image classification in digital pathology, enabling the use of slide-level labels to supervise model training. Although MIL eliminates the tedious fine-grained annotation process for supervised learning, whether it can learn accurate bag- and instance-level classifiers remains a question. To address the issue, instance-level classifiers and instance masks were incorporated to ground the prediction on supporting patches. These methods, while practically improving the performance of MIL methods, may potentially introduce noisy labels. We propose to bridge the gap between commonly used MIL and fully supervised learning by augmenting both the bag- and instance-level learning processes with pseudo-label correction capabilities elicited from weak to strong generalization techniques. The proposed algorithm improves the performance of dual-level MIL algorithms on both bag- and instance-level predictions. Experiments on public pathology datasets showcase the advantage of the proposed methods.
Paper Structure (24 sections, 14 equations, 3 figures, 5 tables)

This paper contains 24 sections, 14 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Formulation of MIL problems
  5. MIL in WSI Classification
  6. Overview of the Model
  7. Bag Branch
  8. Instance Branch
  9. Loss Functions
  10. Scheduling Dual-Level Training
  11. Experiments
  12. Datasets
  13. Evaluation Metrics and Implementation Details
  14. Baselines
  15. Results
  16. ...and 9 more sections

Figures (3)

  • Figure 1: Framework overview of the proposed method.
  • Figure 2: Visualization of instance-level prediction results. (a) shows one original CAMELYON16 slide thumbnail with ground-truth positive instances masked in green. The following are heatmaps overlapped on the slide image where deeper red indicates higher predicted probability of being positive and deeper blue indicates lower probability of being positive.
  • Figure 3: Instance-level AUC on CAMELYON16 test set throughout training.

Theorems & Definitions (1)

  • Remark : A tripartite balance between $L_\text{label}$, $L_\text{inst}$, and $L_\text{self}$