Chiron-o1: Igniting Multimodal Large Language Models towards Generalizable Medical Reasoning via Mentor-Intern Collaborative Search

TL;DR

This paper introduces MICS, a collaborative search scheme for generating high-quality medical reasoning data, leading to a new multimodal medical language model, Chiron-o1, with improved reasoning and visual question-answering abilities.

Contribution

It proposes MICS for effective medical reasoning data generation and develops Chiron-o1, a multimodal medical language model trained on this data, achieving state-of-the-art results.

Findings

Chiron-o1 outperforms existing models on medical VQA benchmarks.

MICS effectively generates high-quality reasoning paths for medical diagnosis.

Chiron-o1 demonstrates robust generalizable medical reasoning capabilities.

Abstract

Multimodal large language models (MLLMs) have begun to demonstrate robust reasoning capabilities on general tasks, yet their application in the medical domain remains in its early stages. Constructing chain-of-thought (CoT) training data is essential for bolstering the reasoning abilities of medical MLLMs. However, existing approaches exhibit a deficiency in offering a comprehensive framework for searching and evaluating effective reasoning paths towards critical diagnosis. To address this challenge, we propose Mentor-Intern Collaborative Search (MICS), a novel reasoning-path searching scheme to generate rigorous and effective medical CoT data. MICS first leverages mentor models to initialize the reasoning, one step at a time, then prompts each intern model to continue the thinking along those initiated paths, and finally selects the optimal reasoning path according to the overall reasoning performance of multiple intern models. The reasoning performance is determined by an MICS-Score, which assesses the quality of generated reasoning paths. Eventually, we construct MMRP, a multi-task medical reasoning dataset with ranked difficulty, and Chiron-o1, a new medical MLLM devised via a curriculum learning strategy, with robust visual question-answering and generalizable reasoning capabilities. Extensive experiments demonstrate that Chiron-o1, trained on our CoT dataset constructed using MICS, achieves state-of-the-art performance across a list of medical visual question answering and reasoning benchmarks. Codes are available at https://github.com/manglu097/Chiron-o1