PathNavigate: Training-Free AI Agent for Pathology Slide VQA

Researchers have introduced PathNavigate, a training-free pathology agent designed for whole-slide image visual question answering (WSI-VQA). The system addresses the challenge of navigating gigapixel pathology slides under strict inspection budgets to answer free-form clinical queries. Unlike existing supervised multimodal large language models (MLLMs) that require retraining, PathNavigate keeps core models frozen. It employs a surprise-guided scan mechanism and shared slide memory to locate relevant high-resolution evidence without relying solely on question-conditioned relevance, which can miss decisive morphology not named in the query. The approach aims to reduce inference-time scaffolding and improve practicality for clinical applications.