Chi-square distance kernel of the gaits for the diagnosis of Parkinson's disease

In this paper, a new approach for the diagnosis of the subjects with Parkinson's disease (PD) from the healthy control subjects is proposed. This method uses the measurements of gait signals using the ground reaction forces under usual walking of the subjects. These measurements were computed using 8 sensors placed underneath of each foot. The absolute value of the difference between the force measurements were calculated for each sensor at each time and these signals went through a short-time Fourier transform (STFT) and several features were extracted from the spectrum of the signals. The histogram of these features was computed and the bin selection was performed using the feature discriminant ratio (FDR) method. Then the chi-square distance between the reduced histograms was computed and it formed a kernel for support vector machines (SVMs) for classification. The results on 93 subjects with PD and 73 healthy control subjects show that the proposed approach obtains an accuracy of 91.20% for the diagnosis of the PD using gait signals.

► A system for the diagnosis of Parkinson's disease using the gait was proposed.
► The force sensors placed underneath of each foot to measure the gait information.
► Features derived using the STFT and their histograms were created.
► The chi-square distance was used for computing the distance between the histograms.
► The SVM with this kernel was used for the diagnosis of Parkinson's disease.