This research demonstrates how generative AI can be trained to detect errors in radiology reports, potentially reducing medical documentation mistakes.

• Trained on 1,656 synthetic chest radiology reports (half with deliberate errors)
• Validated using 614 reports combining real MIMIC-CXR data and synthetic error-containing reports
• Demonstrates how LLMs can be finetuned for specialized medical quality control
• Shows promising potential for AI assistance in clinical workflows

This research matters because radiology report errors can impact patient safety and treatment decisions. AI-powered error detection could serve as an automated second check before reports are finalized.

Generative Large Language Models Trained for Detecting Errors in Radiology Reports