Improved YOLO5 for diagnosis of developmental dysplasia of the hip

Developmental dysplasia of the hip (DDH) refers to the abnormal morphology or position of the femoral head and acetabulum in infants at birth or during the growth and development process due to certain factors, leading to instability of the hip joint, restricted abduction of the infant hip joint, abnormal gait in toddlers, pain and limited mobility during activities in adolescents and adults. Early detection and treatment are crucial in combating this condition. In recent years, several studies have utilized deep learning techniques for diagnosing DDH. However, analyzing hip X-ray images poses challenges due to their complexity and the imbalance in the distribution of DDH data classes, where negative samples outnumber positive ones. This research introduces an enhanced YOLO algorithm designed to address this class imbalance issue by adjusting the loss function and incorporating an attention mechanism. Results indicate that focusing on the specific characteristics of DDH data can enhance model accuracy in DDH detection using deep learning, with the improved YOLO5 model achieving a mAP0.5∼0.95 of 82.5% on the test dataset. Additionally, experiments with networks of varying sizes demonstrate the enhanced algorithm's improved accuracy in diagnosing DDH through deep learning methods. The advancement of artificial intelligence has facilitated the early detection of DDH. The enhanced YOLO model in this study has increased the accuracy of diagnosis, thereby enabling the detection of DDH in situations where professional medical personnel and equipment may be lacking in the future.