1280 Radiology Reports: Original and Error Variants Dataset

This dataset was used in the manuscript titled “Failure of Large Language Models to Detect Clinically Significant Errors in Radiology Reports.” It comprises 256 original radiology reports drawn from CT (n = 104), MRI (n = 104), and X-ray (n = 48) studies spanning multiple anatomical regions. For each original (error-free) report, four error variants (n = 1,024) were created, resulting in a total of 1,280 reports. Each variant introduces one of four predefined clinically significant error types: E1: Anatomical mislabeling that could lead to wrong-site actions. E2: Physiologically impossible or nonsensical findings. E3: Diagnostic inconsistencies that alter staging or diagnosis. E4: Recommendations that contradict the described findings.