MedQPA-Gen: Medical Question Proposing and Answering for Report Generation

Medical report generation from medical im- ages is a vital AI task that helps doctors with diagnosis and marks a significant step toward creating general AI-powered medical systems. However, previous methods either fail to op- timize factual accuracy or heavily depend on expert preference data. To overcome these chal- lenges, we propose MedQPA, an automatic and generalizable report evaluation technique that uses question proposing and answering to enable controllable, structured reasoning grounded in medical domain knowledge and the factual correctness of the report. Addi- tionally, we design MedQPA-Gen, a medical report generation pipeline that maximizes the MedQPA score through prompt engineering and reinforcement learning with MedQPA as a reward signal. We demonstrate that MedQPA is an accurate evaluation metric that closely correlates with human preferences. More importantly, MedQPA-Gen achieves higher human preference scores and better perfor- mance on downstream tasks. We open-source code at this repo https://github.com/MedQPA- gen/MedQPA-gen

Recommended citation: Weijie Liang*, Xiyue Zhu*†, Ruike Zhu, Chenhao Li, Cheng Tang, Zhiyu Liu, Zhihua Gong, Shirui Luo, Yudu Li, Volodymyr Kindratenko.