CAN ARTIFICIAL INTELLIGENCE REPLACE THE ROLE OF THE RADIOLOGIST IN THE REPORTING OF WRIST RADIOGRAPHS

Purpose: Artificial intelligence (AI) has demonstrated improved accuracy and efficiency in several areas of medical imaging; however, its role in musculoskeletal radiograph interpretation remains unclear. This study aimed to evaluate the diagnostic accuracy of readily accessible AI platforms in interpreting wrist radiographs and to determine their ability to detect abnormalities and provide correct diagnoses compared with radiologist reports. Methods: This is a retrospective observational study of 100 consecutive patients who underwent radiographic examination of the wrist referred by the senior author performed at Gleneagles Hospital. A total of 100 anonymised wrist radiographs were included in this study comprising of 50 normal radiographs and 50 abnormal radiographs, with the abnormal images further categorised as trauma-related, degenerative, congenital, or post-operative. Each radiograph was uploaded to two AI platforms (Grok and CT-read) using a standardised prompt requesting a holistic report. AI outputs were assessed for abnormality detection and diagnostic correctness. Sensitivity, specificity, accuracy, positive predictive value (PPV), negative predictive value (NPV), and Youden’s index were calculated and compared against chance performance.