Aligning Findings with Diagnosis: A Self-Consistent Reinforcement Learning Framework for Trustworthy Radiology Reporting

TL;DR

This paper presents a self-consistent reinforcement learning framework for radiology report generation that aligns visual evidence with linguistic output, reducing hallucinations and improving clinical accuracy in multimodal large language models.

Contribution

It introduces a novel 'Reason-then-Summarize' architecture optimized with Group Relative Policy Optimization, explicitly aligning findings with diagnoses and enhancing report reliability.

Findings

Achieves state-of-the-art performance on MIMIC-CXR benchmark.

Significantly reduces hallucinations compared to supervised baselines.

Improves clinical efficacy metrics in radiology report generation.

Abstract

Multimodal Large Language Models (MLLMs) have shown strong potential for radiology report generation, yet their clinical translation is hindered by architectural heterogeneity and the prevalence of factual hallucinations. Standard supervised fine-tuning often fails to strictly align linguistic outputs with visual evidence, while existing reinforcement learning approaches struggle with either prohibitive computational costs or limited exploration. To address these challenges, we propose a comprehensive framework for self-consistent radiology report generation. First, we conduct a systematic evaluation to identify optimal vision encoder and LLM backbone configurations for medical imaging. Building on this foundation, we introduce a novel "Reason-then-Summarize" architecture optimized via Group Relative Policy Optimization (GRPO). This framework restructures generation into two distinct components: a think block for detailed findings and an answer block for structured disease labels. By utilizing a multi-dimensional composite reward function, we explicitly penalize logical discrepancies between the generated narrative and the final diagnosis. Extensive experiments on the MIMIC-CXR benchmark demonstrate that our method achieves state-of-the-art performance in clinical efficacy metrics and significantly reduces hallucinations compared to strong supervised baselines.