Automatic diagnosing of infant hip based on Graf criteria
Xiang Yu, Dongyun Lin, Weiyao Lan, Bingan Zhong, Ping Lv · 2017
In this paper, we proposed an automatic diagnosis method in detection of infants hips, and experimental results on real ultrasonic images have shown its fastness and capability in the eld. Four procedures, pre-processing of raw images, segmenting, feature extracting and diagnosing, are included in proposed method. Pre-processing mainly focus on obtaining interested region from raw images. Segmenting, followed by features extracting from segmented images, proceeded at once after pre-processing. The algorithm of segmentation we used here is region-scalable tting energy model. Finally, we obtain two most important reference indexes of Graf criteria, angles α and β, by tting lines with least squares method applied. Accordingly, hips are classi ed into one of four types, including maturity, dysplasia, severe dysplasia and dislocation, according to aforementioned indexes. Accuracy on practical images reaches 80.4% with 93 images tested.