This research introduces FOCUS, a novel approach that improves cancer diagnosis from whole slide images when limited training data is available.

• Adapts compression using domain knowledge to preserve diagnostically relevant image features
• Implements patch pruning to filter out uninformative tissue regions
• Achieves superior performance across multiple cancer types with fewer labeled examples
• Demonstrates a 13.7% accuracy improvement over previous methods

This innovation addresses critical challenges in computational pathology by enabling more accurate diagnosis with limited data, helping overcome annotation scarcity and patient privacy constraints in medical imaging.

FOCUS: Knowledge-enhanced Adaptive Visual Compression for Few-shot Whole Slide Image Classification