Swing limb detection using a convolutional neural network and a sequential hypothesis test based on foot pressure data during gait initialization in individuals with Parkinson's disease

Objective. Start hesitation is a key issue for individuals with Parkinson's disease (PD) during gait initiation. Visual cues have proven effective in enhancing gait initiation. When applied to laser-light shoes, swing-limb detection efficiently activates the laser on the side of the stance limb, prompting the opposite swing limb to initiate stepping. Approach. This paper presents the development of two models for this purpose: a convolutional neural network that predicts the swing limb's side using center of pressure data, and a swing onset detection model based on sequential hypothesis test using foot pressure data. Main results. Our findings demonstrate an accuracy rate of 85.4% in predicting the swing limb's side, with 82.4% of swing onsets correctly detected within 0.05 s. Significance. This study demonstrates the efficiency of swing-limb detection based on foot pressures. Future research aims to comprehensively assess the impact of this method on improving gait initiation in individuals with PD.