Deepmotion: a deep convolutional neural network on inertial body sensors for gait assessment in multiple sclerosis

Abstract

Walking impairment resulted by various chronic diseases, disorders and injuries have been investigated using recent emerging wearable technology, for instance, gait assessment using inertial body sensors in 6-minute walk (6MW) for persons with Multiple Sclerosis (PwMS) to identify spatiotemporal features useful to assess MS progression. However, most studies to date have investigated the features extracted from movements of the lower limbs and do not provide a holistic gait assessment. A recent pilot study demonstrated that the holistic gait assessment such as evaluating the associations among lower and upper limbs provided better discrimination between healthy controls and PwMS. This paper is motivated by this and further aim to answer the following question: can we identify the temporal gait patterns in terms of the holistic gait assessment? Traditionally this suffers from the statistical property of the causality inference method adopted by previous study. We proposed a deep convolutional neural network (CNN) to learn the temporal and spectral associations among the time-series motion data captured by the inertial body sensors. A simulated model was developed to train the CNN, and then the trained CNN was adopted to assess the gait performance from a pilot dataset with 41 subjects (28 PwMS and 13 healthy controls). Experimental results are reported to illustrate the performance of the proposed approach.