New AI framework enables medical imaging systems to learn new diseases without forgetting previous diagnostic knowledge

Bi-CRCL framework uses dual-learner approach inspired by complementary learning systems to balance retaining old knowledge while adapting to new disease categories

Addresses critical challenge of class-incremental learning in medical imaging where privacy constraints and heterogeneous data prevent traditional memory replay methods

Leverages pretrained foundation models (PFMs) with domain-specific adaptation to handle anatomical complexity and institutional differences in medical datasets

Conservative learner preserves prior diagnostic knowledge through stability-oriented updates while system adapts to emerging conditions