AI Screening System for Hip Dysplasia Achieves 96% Sensitivity on Pelvic Radiographs

An AI system for hip dysplasia screening on pelvic radiographs achieved sensitivity 96%, specificity 92%, and AUC 0.94, outperforming orthopaedic residents (F-measure 0.96 vs 0.60–0.81). Trained on a limited dataset (40 patients), it requires external validation before clinical…

• Retrospective diagnostic accuracy study: 73 hip dysplasia patients (unilateral periacetabular osteotomy) and 29 asymptomatic volunteers; AI trained on 40/validated on 7/tested on 26 dysplasia + 29 volunteers.
• Agreement with ground-truth measurements (hip surgeon) was good to excellent (ICC 0.70–0.95).
• Limitation: single-center, small sample, no external validation, and testing only on surgically confirmed dysplasia cases.

RadPigeon summaries are original and for information only. They are not clinical advice.