LPD: Learnable Prototypes with Diversity Regularization for Weakly Supervised Histopathology Segmentation

2025-12-05

AI Summary

Weakly supervised semantic segmentation (WSSS) in histopathology faces three key challenges: inter-class homogeneity, intra-class heterogeneity, and CAM region shrinkage. Existing two-stage approaches rely on clustering to construct prototype banks, suffering from high computational overhead, sensitivity to hyperparameters, and decoupled prototype learning and segmentation optimization. This paper proposes the first end-to-end learnable prototype framework: it eliminates explicit clustering and jointly optimizes learnable class prototypes and the segmentation head in a single stage; introduces a diversity regularization term to enhance prototype coverage of morphological variations; and achieves pixel-level supervision using only image-level labels. Evaluated on the BCSS-WSSS benchmark, our method establishes new state-of-the-art performance, achieving superior mIoU and mDice scores. Qualitatively, predicted boundaries are sharper and missegmentations are significantly reduced.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Semi-Supervised Learning
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Security and Privacy: Data transparency and provenance

Abstract

Weakly supervised semantic segmentation (WSSS) in histopathology reduces pixel-level labeling by learning from image-level labels, but it is hindered by inter-class homogeneity, intra-class heterogeneity, and CAM-induced region shrinkage (global pooling-based class activation maps whose activations highlight only the most distinctive areas and miss nearby class regions). Recent works address these challenges by constructing a clustering prototype bank and then refining masks in a separate stage; however, such two-stage pipelines are costly, sensitive to hyperparameters, and decouple prototype discovery from segmentation learning, limiting their effectiveness and efficiency. We propose a cluster-free, one-stage learnable-prototype framework with diversity regularization to enhance morphological intra-class heterogeneity coverage. Our approach achieves state-of-the-art (SOTA) performance on BCSS-WSSS, outperforming prior methods in mIoU and mDice. Qualitative segmentation maps show sharper boundaries and fewer mislabels, and activation heatmaps further reveal that, compared with clustering-based prototypes, our learnable prototypes cover more diverse and complementary regions within each class, providing consistent qualitative evidence for their effectiveness.

Problem

• Addresses inter-class homogeneity and intra-class heterogeneity in histopathology segmentation
• Overcomes CAM-induced region shrinkage in weakly supervised semantic segmentation
• Eliminates costly two-stage pipelines by integrating prototype learning with segmentation

Innovation

• Learnable prototypes for one-stage segmentation
• Diversity regularization to enhance intra-class coverage
• Cluster-free framework improving morphological heterogeneity handling