Improving measurement of hip joint center location using neural networks

Abstract

In human movement analysis accuracy of locating the hip joint center (HJC) becomes important in measurements of the hip muscle lengths and hip moment arms. Conventional gait analysis methods use regression and polynomial estimation techniques based on cadaver measurements to locate the HJC. Keeping in view the importance of Neural Networks (NN) in estimation, two Feedforward NN were constructed to estimate the HJC position from training sets of actual HJC positions from MRI data. First network was based on data from 32 subjects (8 adults, 14 children and 10 children with cerebral palsy), and second NN based on 22 healthy subjects. Estimation results were compared with multivariable linear regression (MR) and Newington-Gage (NG) methods. From the validation data, the proposed networks reduced error in HJC position estimation by approximately 69% compared to NG method, and 30% compared to the MR method.