Statistical shape model-based automatic evaluation and validation of anatomical measures on healthy femur shapes from limited field of view imaging

A standard clinical CT scan of the hip joint provides only partial femoral morphometrics due to its limited field-of-view and hence is useless for extracting anatomical measures of the entire femur that may become useful in retrospective studies and applied research, such as improving musculoskeletal models. This study used a statistical shape modelling approach to automatically determine seven anatomical measures of the femur from limited field-of-view 3D imaging. A shape model of the full femur was built from a training sample of 50 adult cadaveric CT images and used to predict anatomical measures from full and simulated limited field-of-view segmentations of the training group and a test group (n = 17). The predictive nature of shape modelling allowed for the extraction of additional distal measures not available in limited field-of-view imaging, with no significant difference between full and limited field-of-view input for both femur groups. The predictions of femoral neck-shaft angles were the most accurate (error < 1%), while the other measure predictions had lower errors ranging between 5% and 10%. The presented pipeline offers a technique to automate anatomical landmark extraction from CT imaging and can robustly predict additional anatomical measures not the original scan, increasing the applicability of limited field of view imaging.