Early Screening of DDH using SVM Classification

Treatment of Developmental Dysplasia of Hip (DDH) becomes less convoluted if it is detected early. In this paper, an acoustic non-invasive data is used for detection of DDH. We investigate early detection of DDH using machine learning technique through support vector machine (SVM) technique. We use data from a proposed method that tested different simplified models of the hip joint. Models were stimulated with band-limited white acoustic noise (10-2500 Hz) and the response of the model was measured. We obtain phase, transfer function and coherence as features for different simulated hip dysplasia levels and for simulated normal cases. Results shows that linear SVM gives an overall accuracy of 79% for 4 class with an area under the curve (AUC) of.93 for the most dislocated hip joint in receiver operating characteristic (ROC) curve.