Reproduce “PathNavigate: A Training-Free Pathology Agent with Surprise-Guided Scan and Shared Slide Memory for Whole-Slide Image VQA” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Chunze Yang, Qidong Liu, Wenjie Zhao, Yue Tang, Jiusong Ge, Di Zhang, Jiashuai Liu, Lei Wu and 5 more

Researchers at Xi’an Jiaotong University developed PATHNAVIGATE, a training-free pathology agent for whole-slide image visual question answering (WSI-VQA) that employs a "Scan-Search-Readout" routine and surprise-guided navigation. This system achieved 59.42% overall accuracy on SlideBench-BCNB and 56.34% total accuracy on WSI-VQA, while reducing prompt token consumption and offline footprint compared to previous training-free methods.

Key insights

  • Establishing a broad, slide-specific content prior before integrating question relevance, mirroring human pathologist behavior, significantly enhances evidence selection and diagnostic accuracy in WSI-VQA.
  • A shared online memory that adapts to scanned regions provides a consistent context for both ROI proposal and answer-time adjudication, improving the system's reasoning capabilities.
  • The synergistic combination of "surprise" (slide atypicality) and question-conditioned relevance is more effective for efficient and targeted high-magnification region selection than either signal alone.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.