MedExpMem: Adapting Experience Memory for Differential Diagnosis

In three linesMedExpMem is an experience memory framework enabling medical vision-language models to accumulate differential diagnosis expertise. Unlike RAG, it memorizes discriminative experiences from past diagnostic failures as pairwise differential notes. Evaluated across 11 radiology subspecialties, it improves accuracy up to 7.0% across diverse models.

Summary generated by Claude — human-verified