Patient-dependent Freezing of Gait Detection using Signals from Multi-accelerometer Sensors in Parkinson’s Disease

Abstract

The position and number of the on-body wearable sensors affects significantly the acquired signal, which sequentially has a direct influence on the patient’s diagnosis. The patients of Parkinson’s disease (PD) suffer from freezing of the gait (FOG) in the form of episodes. In this paper, the choice of the acceleration sensors’ location, which measures the patient’s movement for monitoring the PD patient, was introduced using several episodes to develop a patient-dependent model for FOG detection. The proposed classification using the linear support vector machine (SVM) based FOG detection was applied to the ranked features using infinite feature selection (IFS) method to distinguish between the freezing and no-freezing events. A comparative study between the proposed IFS based detection model and the use of Eigenvector feature selection was conducted showing the same features ranking performance of the extracted features from all acceleration signals from the multi-sensors. However, the results established the superiority of the proposed patient-dependent model using IFS ranked features for FOG detection, which can be used to improve the PD monitoring systems accuracy.