Data

• 940 radiographs from adult patients with pelvic or hip trauma

Design

• Retrospective study

• Junior radiologist analysed the radiographs without and with AI (3-month washout)

• MSK imaging radiologists, junior radiologist and emergency physicians analyzed 10% of the radiographs first without AI, then with AI (100 exams)

• Ground truth established by either CT scan, confirmation by radiography or regular follow-up at the hospital

Results

• 940 patients with pelvic X-rays, 539 patients with at least one fracture

  • Older population with a mean age = 83 years old

• In total, 633 pelvic fractures (64.8% from hip, 35.2% from pelvic ring)

• Good standalone performance of AI across all pelvic fractures (sens = 81%, spec = 78%)

• AI improved the junior radiologist’s sensitivity by 8.39% (Rel Diff) without decreasing specificity

• AI most significantly improved the performance of emergency physicians with an average increase in sensitivity of 29% without any loss of specificity