Clinical study on ultrasonic artificial intelligence-assisted diagnosis of developmental hip dysplasia in children

Background: Developmental hip dysplasia(DDH) is a common pediatric disease.For patients younger than 6 months of age,ultrasound diagnosis is more suitable for screening and assessment of hip development.At present,there is an urgent need for a reproducible and reliable ultrasound screening method for DDH diagnosis. Purpose: To construct and verify an artificial intelligence-assisted deep learning system for ultrasound diagnosis of developmental hip dysplasia in children. Materials and Methods : 2021 standard sections were selected from January 2019 to January 2021. All standard sections were annotated using unified standards through the image transmedia data annotation and audit system.1753 images were randomly selected to train the deep learning system,the remaining 268 were used to test the system. Results: 268 patients were tested. The AUC for diagnosing hip joint maturity was 0.941, (sensitivity 90.5%, specificity 97.8%),while the AUC for Graf classification was 0.685(sensitivity 45.3% specificity 91.7%),compared with clinicians’ measurements. According to the Bland–Altman method, the 95% limits of agreement of α angle was-6.426°~4.811°(Bias=-0.8075,P < 0.001), that of β angle was -5.545°~6.507°(Bias=0.4812,P=0.057). 7 key points measured by AI were statistically different from the clinician values. Conclusions: The artificial intelligence system could quickly and accurately measure the Graf correlation index of standard hip joint ultrasound images.