A Neural Network Approach to Estimate Lower Extremity Muscle Activity during Walking

Abstract

Gait analysis is the study of human locomotion. It plays an essential role in the diagnosis and rehabilitation of gait abnormalities, the study of physiological changes associated with ageing, and the treatment of injuries. Muscle activity is an important gait parameter that controls joint function during walking and provides valuable information about the gait quality. However, current techniques to measure muscle activity, such as electromyogram (EMG) and musculoskeletal modelling tools, have drawbacks. This study develops an artificial neural network (ANN) method to estimate eight lower extremity muscle activities using pelvis, hip, knee and ankle joint angles. It uses an online gait database that contains kinematic and kinetic gait parameters and lower limb EMG. Four training algorithms were explored and investigated. Despite the noticeable differences between the actual and the estimated muscle activities, e.g. gluteus maximus and bicep femoris, the results demonstrate the feasibility of the proposed method in determining the muscle behaviour during walking. The study also shows the potentials of machine learning to compensate for the lack of modality and to provide an insight on the dynamics of muscles in gait. Clinical Relevance- Gait analysis is important in clinical and rehabilitation settings. The proposed method has the potential in reducing the dependency on EMGs and can be an alternative to the musculoskeletal modelling tools in diagnosing, treating, and rehabilitating gait abnormalities.