Random Direct Preference Optimization for Radiography Report Generation

2025-09-19

AI Summary

Current radiology report generation (RRG) methods suffer from insufficient clinical accuracy, hindering real-world deployment. To address this, we propose Randomized Contrastive Sampling (RCS), a model-agnostic preference learning strategy integrated into the Direct Preference Optimization (DPO) framework—requiring neither reward models nor human annotations. RCS automatically constructs high-quality preference pairs from unlabeled data, alleviating dependence on expert-labeled datasets and external reward signals. Coupled with large vision-language model alignment techniques, it enhances clinical consistency in chest X-ray report generation. Evaluated across three state-of-the-art baselines, our approach achieves an average 5% improvement in key clinical metrics—without introducing additional training data. This work establishes a scalable, deployable paradigm for low-resource, high-fidelity RRG, advancing clinical reliability while minimizing annotation and infrastructure overhead.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Knowledge Representation and Reasoning: Preferences
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Radiography Report Generation (RRG) has gained significant attention in medical image analysis as a promising tool for alleviating the growing workload of radiologists. However, despite numerous advancements, existing methods have yet to achieve the quality required for deployment in real-world clinical settings. Meanwhile, large Visual Language Models (VLMs) have demonstrated remarkable progress in the general domain by adopting training strategies originally designed for Large Language Models (LLMs), such as alignment techniques. In this paper, we introduce a model-agnostic framework to enhance RRG accuracy using Direct Preference Optimization (DPO). Our approach leverages random contrastive sampling to construct training pairs, eliminating the need for reward models or human preference annotations. Experiments on supplementing three state-of-the-art models with our Random DPO show that our method improves clinical performance metrics by up to 5%, without requiring any additional training data.

Problem

• Improving radiography report generation accuracy for clinical use
• Optimizing medical image analysis without human preference annotations
• Enhancing clinical metrics through random contrastive sampling technique

Innovation

• Uses Direct Preference Optimization for radiography reports
• Leverages random contrastive sampling for training pairs
• Model-agnostic framework improves clinical metrics without extra data