Pathological Gait Detection of Parkinson's Disease using Sparse Representation

Gait analysis has become an attractive quantitative and non-invasive mechanism that can aid early detection and monitoring of the response of Parkinson's disease sufferers to management schedules. In this paper, we model cycles of human gait as a sparsely represented signal using over-complete dictionary. This representation forms the basis of a classification that allows the recognition of symptomatic subjects. Experiments have been conducted using signals of vertical ground reaction force (GRF) from subjects with Parkinson's disease from the publicly available gait database (physionet.org). Our method achieved a classification accuracy of 83% in recognising pathological cases and represents a significant improvement on previously published results that use a selection of the Fourier transform coefficients as features.