RadFusion: Towards Threshold-Controllable Radiology Report Generation

2026-08-11

AI Summary

Existing radiology report generation models struggle to control the sensitivity–specificity trade-off in diagnostic conclusions and lack support for quantitative validation via ROC curves. This work proposes RadFusion, a novel framework that enables threshold-controllable report generation for the first time. By integrating a multi-label classifier, a visual question answering (VQA)-style generator, and a large language model (LLM)-based rewriting mechanism, RadFusion ensures that diagnostic statements in generated reports strictly adhere to a specified classification threshold while preserving detailed imaging descriptions. Evaluated on MIMIC-CXR, the method faithfully reproduces the classifier’s ROC performance: compared to uncontrolled generation, it achieves a 6.9% sensitivity gain at matched specificity and a 20.7% specificity gain at matched sensitivity, thereby facilitating clinical adaptability and regulatory validation.

Abstract

Automated radiology report generation is advancing rapidly in response to the shortage of radiologists, yet unlike a perception model, existing generation models offer no control over the sensitivity-specificity trade-off of their diagnostic content. Such control is essential because clinical scenarios diverge: emergency triage prioritizes sensitivity to reduce missed findings, whereas confirmatory interpretation emphasizes specificity to limit unnecessary interventions. A single fixed report can neither adapt to these scenarios nor support the ROC-based validation widely expected for regulatory clearance. We introduce RadFusion, a framework that equips report generation with threshold controllability. Our method fuses a multi-label classifier, which provides per-disease confidence scores, with a VQA-based report generator, which describes medical findings in detail; an LLM then rewrites the report so that its stated diagnoses follow the classifier's decisions at the selected threshold while staying grounded in the generator's descriptions. On MIMIC-CXR, the performance of RadFusion conforms to the classifier's ROC curve: sweeping the threshold and mapping the reports back to class labels reproduces the classifier's validated ROC performance. This conformance makes generated reports quantitatively evaluable through ROC analysis, strengthening the case for regulatory clearance, and enables operating-point selection that matches report behavior to clinical context. Moreover, combining the two model types improves diagnostic accuracy over uncontrolled generation: sensitivity increases by 6.9% at matched specificity, and specificity by 20.7% at matched sensitivity. These results show that RadFusion makes report generation clinically adaptable, quantitatively verifiable, and diagnostically more reliable.

Problem

• radiology report generation
• sensitivity-specificity trade-off
• threshold controllability
• clinical adaptability
• ROC analysis

Innovation

• threshold-controllable generation
• radiology report generation
• sensitivity-specificity trade-off
• ROC conformance
• multi-model fusion