A fully automated pipeline of cam-type FAI parameters measurement from clinical computed tomography (CT) images in asymptomatic patients

A fully automated pipeline, CAM-FAI-CAD, was used to automatically obtain slice selection, segmentation, landmark detection, and measurement of parameters related to femoral deformity to establish an auxiliary measurement algorithm for developmental screening or diagnosis of cam-type FAI using clinical computed abdomen tomography (CT) images of asymptomatic patients. The novel CAM-FAI-CAD pipeline consists of two components: (i) the use of a convolutional neural network (CNN) based on Le-Net to generate accurate slice selection to identify hip in clinical CT images(ii) the use of image processing algorithms for diagnostic landmark detection to quantify cam-type FAI-related parameters in asymptomatic patients. The CAM-FAI-CAD pipeline was used to train CNN for hip slice selection (n=14500 slices) and analyze clinical CT images of the femur to measure the alpha angle (AA) and femoral head-neck offset (FHNO) in male (n=17) and female (n=13) asymptomatic patients. Automatic CAM -FAI-CAD results were obtained for slice selection, segmentation, and center of femoral head detection as demonstrated by the accuracy (98.5%), Dice similarity index (DSI; 0.9 in the axial plane and 0.92 in the coronal plane) and accuracy (94% in the axial plane and 0.96 in the coronal plane) respectively, while the accuracy of the automatically acquired parameters for AA and FHNO was both around 88% compared to the radiologist’s manually calculated parameters. AA and FHNO measurements using the fully automated CAM-FAI-CAD pipeline were highly in agreement with radiologists’ measurements and could be used to assess cam-type FAI with high diagnostic accuracy and efficiency or evolutionary.