Medical Decision Tree-Enhanced LLMs for Interpretable Reasoning

Ruoyu Liu, Xiaofan Zhang, Shaoting Zhang

Keywords: retrieval-augmented generation, medical decision tree, medical reasoning

TL;DR: Using medical decision trees as the retrieval source for RAG systems in medical reasoning scenarios.

Abstract: Large Language Models have made significant strides in medical reasoning. However, challenges remain due to their limited medical knowledge and the risk of hallucinations. While RAG methods can mitigate these issues by retrieving relevant medical information, they typically supply verbose text fragments, which challenges the model's comprehension. Inspired by the widespread use and inherent interpretability of medical decision trees in clinical practice, we propose Medical Decision Tree RAG (MDT-RAG), a novel RAG framework specifically designed for medical reasoning. In this approach, clinical guidelines containing diagnostic and therapeutic information are first converted into decision trees, which are then used to augment LLMs in place of raw text. Experiments demonstrate that our method not only enhances the performance of medical LLMs in reasoning tasks but also exhibits strong interpretability. All related resources have been made publicly available.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 10648