SegMix: Shuffle-Based Feedback Learning Method for Pathology Image Segmentation

A new shuffle-based feedback learning method called SegMix has been proposed to address limitations in semantic segmentation of pathology images. The approach is inspired by curriculum learning and aims to overcome challenges in acquiring high-quality pixel-level supervised segmentation data, which requires significant workload from experienced pathologists and limits deep learning applications. By relaxing label conditions to image-level classification labels, more data can be utilized across various scenarios. Current methods leveraging Class Activation Map (CAM) to generate pseudo pixel-level annotations from image-level labels fail to thoroughly explore essential characteristics of pathology images, identifying only small areas insufficient for pseudo masking. Segmentation remains critical in computational pathology for identifying disease-affected or abnormal growth areas essential for diagnosis and treatment. The research is documented in arXiv:2604.15777v1 as a cross announcement.