Automatic segmentation and classification of neonatal hips according to Graf’s sonographic method: A computer-aided diagnosis system

Graf’s technique for hip ultrasonography (US) assessment is a well-known and universally accepted method for the evaluation of neonatal hips. However, the training process is long and requires supervision until the evaluator achieves expertise. Computer-based segmentation results may be helpful to less experienced evaluators in the discrimination of anatomical structures and the classification of hip US images. The aim of this research was to construct a fully automatic computer-aided diagnosis (CAD) system for the classification of newborn hip US images to be used as a guide in the training and evaluation processes. The proposal included a cascaded framework that utilizes particle swarm optimization (PSO) and statistical level set (SLS) method of segmentation. The location of the initial contour and the region of interest (ROI) were determined using the PSO method. The SLS method was applied in the ROI in order to segment the critical anatomical structures with great success. The proposed tool utilized the knowledge of these anatomical structures to draw lines and define the alpha and beta angles. The specificity rate of the proposed system in the classification of 164 randomly selected hip ultrasound images was 98.57%. The proposed CAD system is a very promising tool for the segmentation and classification of neonatal hip US images according to Graf’s basic types: normal (Type I), mild dysplasia (Type IIa, IIb) and severe dysplasia (Type IIc, D).