Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dr. Assistant: Enhancing Clinical Diagnostic Inquiry via Structured Diagnostic Reasoning Data and Reinforcement Learning

Yue Guo, Fanfu Wang, Jianwei Lv, Xincheng Shi, Yuchen Li, Youya Wang, Yunsheng Zeng, Yujing Liu, Yunhao Qiao, Gen Li, Junfeng Wang, Bo Yuan

TL;DR

This work tackles the limitations of traditional CDSSs and clinical LLMs in diagnostic inquiry by introducing Clinical Diagnostic Reasoning Data (CDRD), which structurally encodes symptom→evidence→differentials. It then presents Dr. Assistant, a diagnostic model trained with supervised fine-tuning on QA generated from CDRD and refined via reinforcement learning that optimizes diagnostic reasoning and multi-turn inquiry while enforcing fidelity to the CDRD. A dedicated benchmark with real cases and inquiry dialogues evaluates ICD-10 matching and physician satisfaction, demonstrating that Dr. Assistant outperforms open-source rivals and approaches closed-source leaders. The approach offers a practical path to robust, guideline-aligned diagnostic inquiry guidance in clinical decision support systems.

Abstract

Clinical Decision Support Systems (CDSSs) provide reasoning and inquiry guidance for physicians, yet they face notable challenges, including high maintenance costs and low generalization capability. Recently, Large Language Models (LLMs) have been widely adopted in healthcare due to their extensive knowledge reserves, retrieval, and communication capabilities. While LLMs show promise and excel at medical benchmarks, their diagnostic reasoning and inquiry skills are constrained. To mitigate this issue, we propose (1) Clinical Diagnostic Reasoning Data (CDRD) structure to capture abstract clinical reasoning logic, and a pipeline for its construction, and (2) the Dr. Assistant, a clinical diagnostic model equipped with clinical reasoning and inquiry skills. Its training involves a two-stage process: SFT, followed by RL with a tailored reward function. We also introduce a benchmark to evaluate both diagnostic reasoning and inquiry. Our experiments demonstrate that the Dr. Assistant outperforms open-source models and achieves competitive performance to closed-source models, providing an effective solution for clinical diagnostic inquiry guidance.

Dr. Assistant: Enhancing Clinical Diagnostic Inquiry via Structured Diagnostic Reasoning Data and Reinforcement Learning

TL;DR

This work tackles the limitations of traditional CDSSs and clinical LLMs in diagnostic inquiry by introducing Clinical Diagnostic Reasoning Data (CDRD), which structurally encodes symptom→evidence→differentials. It then presents Dr. Assistant, a diagnostic model trained with supervised fine-tuning on QA generated from CDRD and refined via reinforcement learning that optimizes diagnostic reasoning and multi-turn inquiry while enforcing fidelity to the CDRD. A dedicated benchmark with real cases and inquiry dialogues evaluates ICD-10 matching and physician satisfaction, demonstrating that Dr. Assistant outperforms open-source rivals and approaches closed-source leaders. The approach offers a practical path to robust, guideline-aligned diagnostic inquiry guidance in clinical decision support systems.

Abstract

Clinical Decision Support Systems (CDSSs) provide reasoning and inquiry guidance for physicians, yet they face notable challenges, including high maintenance costs and low generalization capability. Recently, Large Language Models (LLMs) have been widely adopted in healthcare due to their extensive knowledge reserves, retrieval, and communication capabilities. While LLMs show promise and excel at medical benchmarks, their diagnostic reasoning and inquiry skills are constrained. To mitigate this issue, we propose (1) Clinical Diagnostic Reasoning Data (CDRD) structure to capture abstract clinical reasoning logic, and a pipeline for its construction, and (2) the Dr. Assistant, a clinical diagnostic model equipped with clinical reasoning and inquiry skills. Its training involves a two-stage process: SFT, followed by RL with a tailored reward function. We also introduce a benchmark to evaluate both diagnostic reasoning and inquiry. Our experiments demonstrate that the Dr. Assistant outperforms open-source models and achieves competitive performance to closed-source models, providing an effective solution for clinical diagnostic inquiry guidance.
Paper Structure (39 sections, 16 equations, 6 figures, 6 tables, 1 algorithm)

This paper contains 39 sections, 16 equations, 6 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Data Construction
  4. CDRD Format for Abstract Logic
  5. CDRD Construction Pipeline
  6. QA Pair for SFT
  7. Multi-turn Inquiry for RL
  8. Dr. Assistant
  9. SFT with $\boldsymbol{P}_{\text{SFT}}$.
  10. RL with multi-turn inquiry.
  11. Experiments
  12. Experimental Settings
  13. ICD Matching for Diagnostic Reasoning
  14. Physician Satisfaction
  15. General Medical Knowledge Retention
  16. ...and 24 more sections

Figures (6)

  • Figure 1: Dr. Assistant provides guidance (b) to physicians in the reasoning-inquiry loop (a), where diagnostic hypotheses both drive and are refined by ongoing inquiry. Since diagnostic hypotheses are central to reasoning, they are our primary focus. We detail (a) in Appendix \ref{['app:loop']}.
  • Figure 2: The scattered logic in clinical guideline.
  • Figure 3: Our workflow involves CDRD construction (a), data synthesis (b, c), and Dr. Assistant training (d).
  • Figure 4: Data distribution of our benchmark. One includes 242 patient profiles across 7 sub-departments (a). The other includes 147 rounds of clinical inquiry dialogues (b). "General Internal Medicine" means the inquiries cover multiple sub-departments.
  • Figure 5: Physician's satisfaction. For a clearer display of comparison, tie cases are excluded from this figure.
  • ...and 1 more figures