Precision Recall Controllable Radiology Report Generation via Hybrid Natural Language and Clinical Reward Learning

The paper presents a novel reinforcement learning framework for automated radiology report generation (RRG) that enables precision and recall control during report generation. By introducing a clinical reward mechanism and a group-relative training strategy, the model improves clinical efficacy while maintaining natural language generation quality, as demonstrated by superior performance on the MIMIC-CXR dataset compared to state-of-the-art methods. This approach is significant for practitioners as it allows for tailored report generation that meets specific clinical requirements, enhancing the applicability of AI in medical contexts.