PathoSCOPE: Few-Shot Pathology Detection via Self-Supervised Contrastive Learning and Pathology-Informed Synthetic Embeddings

2025-05-23

AI Summary

To address the challenge of constructing pathology detection models under extreme scarcity of healthy samples (as few as two) and stringent privacy regulations in medical imaging, this paper proposes an unsupervised unknown pathology detection framework requiring only a minimal number of non-pathological samples. Methodologically, we introduce a Global-Local Contrastive Loss (GLCL) to enhance discriminability of healthy representations and a Pathology-guided synthetic Embedding Generation (PiEG) module for lightweight, pathology-aware feature synthesis. Leveraging self-supervised contrastive learning with an efficient encoder, our approach achieves state-of-the-art unsupervised performance on BraTS2020 and ChestX-ray8, with only 2.48 GFLOPs computational cost and inference speed of 166 FPS. Our key contribution is the first demonstration of reliable unknown pathology localization and generalization under ultra-scarce healthy-sample regimes—simultaneously ensuring data efficiency, privacy compliance, and clinical practicality.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: Bias, Fairness & Privacy
• Search and Optimization: Learning to Search

Application Category

• Security and Privacy: Data transparency and provenance
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Unsupervised pathology detection trains models on non-pathological data to flag deviations as pathologies, offering strong generalizability for identifying novel diseases and avoiding costly annotations. However, building reliable normality models requires vast healthy datasets, as hospitals' data is inherently biased toward symptomatic populations, while privacy regulations hinder the assembly of representative healthy cohorts. To address this limitation, we propose PathoSCOPE, a few-shot unsupervised pathology detection framework that requires only a small set of non-pathological samples (minimum 2 shots), significantly improving data efficiency. We introduce Global-Local Contrastive Loss (GLCL), comprised of a Local Contrastive Loss to reduce the variability of non-pathological embeddings and a Global Contrastive Loss to enhance the discrimination of pathological regions. We also propose a Pathology-informed Embedding Generation (PiEG) module that synthesizes pathological embeddings guided by the global loss, better exploiting the limited non-pathological samples. Evaluated on the BraTS2020 and ChestXray8 datasets, PathoSCOPE achieves state-of-the-art performance among unsupervised methods while maintaining computational efficiency (2.48 GFLOPs, 166 FPS).

Problem

• Detect pathologies with few healthy samples using self-supervised learning
• Reduce reliance on large healthy datasets via synthetic embeddings
• Improve pathology detection accuracy while maintaining computational efficiency

Innovation

• Few-shot pathology detection with minimal healthy samples
• Global-Local Contrastive Loss for embedding optimization
• Pathology-informed synthetic embeddings generation module