Abstract:
The location of the hip joint centre (HJC) is critical for accurate lower limb kinematics. A number of methods allow the HJC to be predicted from the locations of bony pelvic landmarks. However, widely used predictions methods are often developed on small populations, or have inappropriate parameters when considering different populations. We compare the accuracy of prediction methods by Tylkowski[1], Bell[2], and Seidel[3], and update their parameters using a large urban population. 3-D models of the pelvis were automatically segmented from 159 (86 male, 73 female) post-mortem CT scans collected at the Victorian Insitute of Forensic Medicine. The dataset reflects a contemporary western urban adult population from the state of Victoria, Australia. Bony landmarks (ASIS, PSIS, symphysis pubis) were defined on an atlas model and propagated to correspondent positions on each subject-specific model. The three published methods above were used to predict HJC locations first using their published parameters, then using parameters fitted to the current dataset. Ground truth HJC locations were calculated as the centre of a sphere fitted to the acetabular regions of each model. Using published parameters, mean errors in millimetres for the Tylkowski, Bell, and Seidel methods were, respectively, 23 (4.9), 26 (4.1), and 18 (3.9). After fitting parameters to the current dataset, corresponding mean errors were 13 (5.5), 7.3(4.0), and 5.7 (3.3). Published parameter errors were similar to published errors for the Tylkowski and Bell methods, and more than twice that published for the Seidel method. After fitting parameters, errors for all methods were significantly lower than those previously published. These results highlight the need to validate and recalibrate joint centre prediction methods on large and population-specific datasets.