A Neural Network Model for Estimation of Net Joint Moments During Normal Gait

The purpose of this study was to develop a model to estimate lower extremity joint moments during level gait. A three-layer artificial neural network was developed to map diverse inputs (demographics, anthropometrics, electromyography, kinematics) onto sagittal plane resultant joint moments for a sample of healthy young adults. Overall model performance and prediction accuracy were acceptable for the hip, knee, and ankle, with coefficients of determination (r 2 ) reaching 0.90 for the hip and knee and 0.95 for the ankle. Estimates in the case-specific validation produced r 2 values of 0.95, 0.94, and 0.99 for the hip, knee, and ankle, respectively. Absolute errors of estimation for peak data were within the ranges published previously for other joint moment models. The results indicated that the model used in this study is accurate in estimating sagittal plane joint moments about the hip, knee, and ankle. Furthermore, the model retained accuracy with a reduced list of inputs (kinematics and demographics). Future development will include clinical samples to determine the adaptability of this model to the diverse conditions of musculoskeletal gait dysfunction common in the clinical setting.