An AI algorithm for 3D ultrasound correctly categorized 90% of dysplastic hips and 86% of normal/borderline hips, showing high agreement (kappa = 0.79) with clinical 2D ultrasound diagnosis.
Observational (n=1,563)
Yes
Does an AI algorithm analyzing 3D ultrasound accurately diagnose developmental dysplasia of the hip compared to conventional 2D ultrasound clinical diagnosis in infants?
An FDA-cleared AI algorithm analyzing 3D ultrasound images demonstrated high diagnostic accuracy and agreement with human expert clinical diagnosis using 2D ultrasound for developmental dysplasia of the hip in infants.
Effect estimate: kappa 0.79
Three-dimensional (3D) Ultrasonography of the hip introduces opportunities for automation by artificial intelligence (AI) to diagnose developmental dysplasia of the hip (DDH). We compared conventional clinical diagnosis and a commercial FDA-cleared AI algorithm in detecting and classifying hip dysplasia on 3DUS. To validate the accuracy of AI in detection and classification of DDH from 3D US. Infant hip 3D US obtained prospectively for clinical suspicion of DDH at two tertiary pediatric hospitals (in Edmonton, Canada and Melbourne, Australia) from January 2012 to December 2020 were analyzed retrospectively by an AI algorithm trained on other data from Edmonton and not previously exposed to Melbourne images. Results were compared to the clinical diagnosis made using concurrently obtained conventional two-dimensional (2D) US of the same hips. The AI algorithm automatically calculated two indices (alpha angle and femoral head coverage) without user input and classified hips as normal or dysplastic. Diagnostic accuracy of AI for human expert reference-standard diagnosis, and inter-rater reliability between AI and clinical diagnosis were assessed. 2492 hips from 1563 patients were evaluated (1294 hips in Edmonton, 1198 in Melbourne; 1061 (68%) female patients, mean age 87 days, range 4–267 days). 2327/2492 hips were clinically classified as Normal or Borderline (initially Graf IIa but spontaneously normalizing at follow-up imaging), while 165 hips were clinically Dysplastic and treated. AI correctly categorized 90% (148/165) of Dysplastic hips and 86% (2006/2327) of the other hips overall, with sensitivity 0.90 for dysplasia. Only one case of severe dysplasia (Graf III) was missed by AI. Agreement between AI and clinical diagnosis was very high (kappa = 0.79). Automated artificial intelligence analysis of 3D ultrasound images correlated well to human expert clinical diagnosis of hip dysplasia from concurrent conventional 2D ultrasound of the same hips.
Ghasseminia et al. (Sat,) conducted a observational in Developmental dysplasia of the hip (DDH) (n=1,563). FDA-cleared AI algorithm on 3D ultrasound vs. Clinical diagnosis using conventional 2D ultrasound was evaluated on Diagnostic accuracy and inter-rater reliability between AI and clinical diagnosis (kappa 0.79). An AI algorithm for 3D ultrasound correctly categorized 90% of dysplastic hips and 86% of normal/borderline hips, showing high agreement (kappa = 0.79) with clinical 2D ultrasound diagnosis.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: