Joint moment estimation from electromyography of patients with osteoarthritis

Abstract

Biomechanical gait analysis may be used to determine treatment options, evaluate the success of rehabilitation programs or post-surgery recuperation, and provide insight for surgical planning, including functional outcomes for patients. However, gait analysis requires expensive equipment - a limiting factor for many clinical settings. One alternative that has been examined is the utilization of an artificial neural network (ANN) to model nonlinear relationships of gait. Researchers have shown initial success in ANN predictions of pathological conditions in gait as well as modeling other parameters. The purpose of this study was to evaluate the performance of a previously developed three layer feed-forward ANN model at estimating ankle, knee and hip joint moments for subjects with osteoarthrits (OA) from surface electromyography (EMG) signals. The broader purpose was to further validate the use of the ANN model as an alternative, less expensive method to traditional gait analysis. Eighteen subjects (13 female, 5 male) with physician diagnosed OA participated in this study. Each subject completed a full gait analysis session.

Data from surface EMGs located on seven muscles of their symptomatic lower limb was recorded as well as kinematic and ground reaction force data. After post-processing, this data was entered into the ANN model and the model's ability to estimate lower extremity joint moments was evaluated. The ANN was able to accurately map the joint moments of subjects with OA. For the ankle, the mean correlation coefficients between the experimental joint moment and the estimated joint moment were between 0.97 and > 0.99. For the knee, the values ranged between 0.89-0.97, and for the hip, the values were between 0.91 and 0.97. Additionally, the ANN's ability to accurately map the lower extremity joint moments was demonstrated through the evaluation of case-specific joint moment time-series curves. Results have been presented in this study to show that the ANN model is adaptive to a subject group with more diversity than the previously tested group of young, healthy subjects. These findings further support the supposition that the ANN model may provide an alternative to traditional gait analysis methods.