Hip joint center (HJC) is frequently used for determining the femoral anatomical reference frame that was necessary for deriving hindlimb kinetics and kinematics in gait analysis. Although medical images were suggested to obtain accurate HJC position, it’s not routinely performed due to several shortcomings. Numerous predictive methods were proposed but their performances in canine HJC estimation are unclear. Therefore, the objective of this study is to apply the existing predictive regression models utilized in human HJC localization and a nonlinear random forest regression to estimate canine HJC position. The CT scans of 22 healthy mature Taiwan dogs were taken to obtain morphometric data of the pelvis and femur. Three linear regression models (Bell’s, Harrington’s, and Hara’s methods) and a random forest regression were established. Leave-one-out cross-validation was employed to construct model coefficients and assess the prediction errors of the HJC position. The results showed that no differences in Euclidean distances between predictive and true HJC positions were found among prediction models, but the accuracy in cranial/caudal direction obtained from Harrington’s model was superior to the Bell’s model. With the prediction errors less than 5.1 mm, the predictive models appeared to be promising for estimating the canine HJC position. However, whether the current findings can be extrapolated to pathological populations or breeds with vast differences in body conformation are unclear. The present study explored the potentials of predictive models for estimating canine HJC positions, contributing to a more technically feasible means for canine gait analysis.
