PathCoT: Chain-of-Thought Prompting for Zero-shot Pathology Visual Reasoning

Junjie Zhou, Yingli Zuo, Shichang Feng, Peng Wan, Qi Zhu, Daoqiang Zhang *, Wei Shao * · 2025

Read PathCoT: Chain-of-Thought Prompting for Zero-shot Pathology Visual Reasoning by Junjie Zhou, Yingli Zuo, Shichang Feng, Peng Wan, Qi Zhu, Daoqiang Zhang *, Wei Shao * (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-01029.txt.

Description

Junjie Zhou and co-authors present PathCoT, a 2025 prompting method that lets a multimodal language model reason over pathology images step by step, guided by expert knowledge. General models stumble on tissue images: they lack pathology knowledge and can hallucinate findings. Stepwise reasoning helps, but each extra step can also wander off track.

PathCoT steers the reasoning with prior expert knowledge, so the model walks the image like a pathologist would. It needs no task specific training data, working zero shot. A self evaluation step checks the chain and curbs answer drift. The design answers the two failure modes head on: ignorance, supplied by the knowledge prompts, and divergence, curbed by checking itself. Reliable visual reasoning in pathology could support diagnosis in places short of specialists. Methods that ask the model to show its work also let a human audit the work.

Catalog record

CREATORJunjie Zhou, Yingli Zuo, Shichang Feng, Peng Wan, Qi Zhu, Daoqiang Zhang *, Wei Shao *
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE21 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-01029

Browse the full library · About the Tower of Babel