---
title: "PathAgentBench: Benchmarking Evidence-Seeking Vision-Language Models on Whole-Slide Pathology Image"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.19261v1"
permalink: "https://feed.astralor.com/items/itm_fc4e40d6531ca5e07f4d9264"
published: "2026-07-21T16:33:44.000Z"
score: 74
channel: "finance"
tags: ["PathAgentBench","Whole-Slide Pathology Image","Multi-Scale Reasoning","Pathology Vision-Language Model","Evidence Acquisition","Diagnostic Region Localization"]
---

# PathAgentBench: Benchmarking Evidence-Seeking Vision-Language Models on Whole-Slide Pathology Image

## AI commentary

> 这篇论文最值得看的地方在于它没有重复“模型在裁剪好的图块上表现多好”的老故事，而是直面病理AI最核心的痛点：模型到底能不能像医生一样在整张切片中找到病变？答案让人清醒——定位和探索能力几乎为零。这对所有从事医学影像AI的研究者都是一个重要提醒：别光盯着已经切好的数据沾沾自喜，真正的挑战在真实的“寻找”过程中。对开发病理辅助诊断产品的团队来说，这个基准是评估自家模型能力的必备工具。

## At a glance

PathAgentBench is a benchmark specifically designed to evaluate evidence-seeking capabilities of vision-language models on whole-slide pathology images (WSIs). It covers four complementary abilities: cross-modal matching, diagnostic-region localization, multi-scale reasoning, and autonomous exploration. The benchmark includes 1,822 TCGA slides and 17,135 diagnostic paths annotated by ten certified pathologists. Evaluating 20 models reveals strong performance in multi-scale reasoning (>93% accuracy) but extremely poor results in region localization (IoU <0.09) and autonomous exploration (hit rate 0.020 at high magnification), exposing a pronounced gap between reasoning and evidence acquisition.

## Key points

- PathAgentBench builds a diagnostic tree structure linking nested regions across magnifications with scale-specific findings, forming a unified evaluation framework.
- Even the best models achieve an IoU below 0.09 on diagnostic-region localization, underperforming a simple center-based heuristic.
- During autonomous exploration, the unconditional hit rate drops to 0.020 at high magnification, indicating models struggle to acquire evidence directly from gigapixel images.
- The benchmark releases large-scale annotated data and standardized evaluation protocols, pushing pathology AI from 'evidence reasoning' to 'evidence acquisition'.

## Full text

Whole-slide image (WSI) diagnosis requires identifying diagnostically relevant regions, examining them across magnifications, and integrating multi-scale evidence. However, most existing pathology benchmarks evaluate models on pre-cropped patches or pre-extracted slide features, leaving their ability to acquire evidence directly from gigapixel WSIs largely untested. We introduce PathAgentBench, a benchmark for evaluating evidence-seeking vision-language models (VLMs) across four complementary capabilities: image-to-text matching for evidence interpretation, text-to-image retrieval for evidence verification, diagnostic-region localization for evidence acquisition, and multi-scale reasoning for evidence integration. The benchmark is organized as a diagnostic tree that links nested regions across magnifications with scale-specific findings and path-level diagnoses. It contains 1,822 TCGA WSIs and 17,135 diagnostic paths annotated by ten board-certified pathologists. An additional private cohort of 190 breast cancer WSIs with detailed annotations is used to evaluate autonomous whole-slide exploration. We evaluate 20 general-purpose, medical, and pathology-specialized models. Leading open-weight models achieve over 93% accuracy in multi-scale reasoning and over 50% accuracy in both cross-modal matching tasks. In contrast, diagnostic-region localization remains challenging: the best text-guided mean intersection-over-union is below 0.09, underperforming a simple center-based heuristic. During autonomous exploration, the unconditional hit rate decreases from 0.522 at low magnification to 0.185 at intermediate magnification and 0.020 at high magnification. These results reveal a pronounced gap between reasoning over curated evidence and acquiring that evidence directly from WSIs. PathAgentBench provides a unified framework for measuring and improving evidence-seeking pathology models.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_fc4e40d6531ca5e07f4d9264