Femoral neck-shaft angle (NSA) is the angle included by the femoral neck axis (FNA) and the femoral shaft axis (FSA), which is a critical anatomic measurement index for evaluating the biomechanics of the hip joint. Aiming at solving the problem that the physician’s manual measurement of the NSA is time consuming and irreproducible, this paper proposes a fully automatic approach for evaluating the femoral NSA on radiographs. We first present an improved deep convolutional generative adversarial network to automatically segment the femoral region of interest on radiographs of the pelvis. Then based on the geometrical characteristic of the femoral shape, the FNA and FSA are fitted, respectively, and thus, the NSA can be evaluated conveniently. The average accuracy of the proposed approach for NSA evaluation is 97.24%, and the average deviation is 2.58° as compared to the measurements manually evaluated by experienced physicians. There is no significant statistical difference (P = 0.808) between the manual and automatic measurements, and Pearson’s correlation coefficient is 0.904. It is validated that the proposed approach can provide an effective and reliable tool for automatically evaluating the NSA on radiographs.

No abstract

No abstract

No abstract

No abstract