Enhancing Medical Students’ Disease Detection Skills: Leveraging Artificial Intelligence Software for Diagnosing Developmental Dysplasia Using Ultrasound Imaging

Background: Ultrasound is the gold standard for the screening of developmental dysplasia of the hip (DDH). However, there are several problems in overcoming the learning curve of technical skills required to detect DDH using ultrasound. Thus, several attempts have been made to assist in the detection of DDH using artificial intelligence (AI) software (SW). This study aimed to evaluate the efficacy of AI in DDH ultrasound screening to aid medical students in the diagnosis of DDH. Method: A Mask R-CNN-based detection model was developed to identify six key points and measure the alpha and beta angles, as well as containment, following Graf’s method. Fifty-three second-year medical students assessed 38 DDH screening ultrasound images from 28 patients twice: first through self-measurement after brief training and then using AI-assisted measurement, where key points were refined based on AI predictions. The gold standard for comparison was the measurement performed by an experienced orthopedic surgeon. Results: AI assistance significantly improved agreement with the expert reference for all three measurements. The proportion of students classified in the excellent agreement category increased from 26.4% to 81.1% for the alpha angle, from 67.9% to 86.8% for the beta angle, and from 56.6% to 84.9% for containment. The mean ICC increased from 0.593 to 0.855 for the alpha angle, from 0.760 to 0.892 for the beta angle, and from 0.722 to 0.882 for containment. Wilcoxon signed-rank tests demonstrated significant improvements in all three measurements (all p < 0.001). Conclusions: AI-assisted ultrasound measurement may improve agreement with expert measurements among novice users assessing DDH. This approach may have potential as a supportive tool for DDH screening and educational training by reducing the learning curve associated with ultrasound-based measurement.