PDD-RRG: Posterior Diagnostic Decision for Study-level Radiology Report Generation

2026-08-03

AI Summary

Existing radiology report generation methods struggle to effectively integrate multi-view images and historical clinical information, often introducing diagnostic bias. This work proposes a novel posterior diagnostic decision mechanism that, after generating initial multi-view reports, refines the final output by fusing diagnostic conclusions through Bayesian posterior probabilities and learned clinical observation thresholds. Notably, this approach enhances clinical validity without requiring model retraining and represents the first integration of posterior diagnostic reasoning into the report generation pipeline. Evaluated on the MIMIC-CXR dataset, the method significantly improves both diagnostic accuracy and overall report quality of existing radiology report generation models.

Abstract

Automatic radiology report generation (RRG) aims to simulate the workflow of radiologists, assisting them in clinical diagnosis. However, existing methods often fall short in utilizing all information relevant to the examination, as is typically done in clinical practice. Although some works attempt to incorporate multi-view images and historical data, these additional inputs may sometimes lead to avoidable diagnostic errors on the contrary. To address these challenges, we introduce a decision-making stage after report generation for the first time and propose a Posterior Diagnostic Decision framework (PDD-RRG) to integrate potentially conflicting diagnoses. Specifically, we create various subsets of input data and utilize an existing RRG model to generate reports from different perspectives. Then the Bayesian posterior probability and the learned thresholds for each clinical observation are calculated to obtain an aggregated diagnostic conclusion, which is subsequently used to refine the generated report. Experiments on MIMIC-CXR demonstrate that our proposed PDD-RRG can effectively enhance the clinical efficacy of existing RRG models without any retraining.

Problem

• radiology report generation
• diagnostic errors
• clinical diagnosis
• multi-view images
• historical data

Innovation

• Posterior Diagnostic Decision
• Radiology Report Generation
• Bayesian Posterior Probability
• Multi-perspective Diagnosis Fusion
• Clinical Efficacy Enhancement