[arXiv]score: 0.49
PathMem Adds Structured Long-Term Memory to Pathology MLLMs for Diagnostic Reasoning
May 26, 2026
PathMem introduces a Memory Transformer that dynamically transitions structured pathology knowledge (taxonomy, grading criteria) from long-term to working memory in MLLMs, improving consistency with diagnostic standards.
cs.AI
HOW THIS AFFECTS YOU
●
researcherThe hierarchical LTM-to-working-memory architecture offers a transferable pattern for injecting structured domain knowledge into MLLMs beyond pathology.
●
healthWorth watching because explicit memory control for pathology-specific grading criteria addresses a key reliability gap in computational pathology MLLMs used in diagnostic workflows.