This research advances weakly supervised classification of medical whole slide images (WSIs) when training data is limited, particularly for rare diseases.

• Introduces Multi-scale and Context-focused Prompt Tuning (MSCPT) to enhance few-shot learning capabilities
• Leverages pre-trained Vision-Language Models to extract meaningful features from pathology slides
• Addresses the critical challenge of rare disease identification with minimal labeled examples
• Demonstrates a practical approach to reducing the data requirements for digital pathology systems

This matters for healthcare applications by enabling more accurate diagnosis with fewer training samples, potentially accelerating deployment of AI-assisted pathology tools for rare conditions.

MSCPT: Few-shot Whole Slide Image Classification with Multi-scale and Context-focused Prompt Tuning