Distinguishing between Parkinson’s disease patients and healthy individuals using a comprehensive set of time, frequency and time-frequency features extracted from vertical ground reaction force data

Abstract

Parkinson’s disease (PD) is a progressive neurodegenerative disease manifests by motor and non-motor symptoms. During disease progression, several movement disorders appear that influence subject’s natural life. Diagnosis of PD, especially at early-stages is important for early medication and other related interventions. In this regard, intelligent systems are interesting opportunities for PD diagnosis. In this study, a classification method for discriminating PD patients from healthy individuals was proposed in which using several feature sets extracted from vertical ground reaction force (VGRF) data and incorporating a decision tree classifier, higher classification performance was obtained compared with other existing methods. The feature sets were extracted for time, frequency and time-frequency domains by considering both local and global attributes of dynamic characteristics of human walking. The obtained results showed that considering features from different domains (time, frequency and time-frequency) enhanced classification performance in a large extent. Also, the statistical analysis of extracted features showed that PD patients performed the stance phase of the gait cycle at a delayed and prolonged duration, with an increased total applied force compared with the healthy group. Furthermore, the VGRF frequency content showed skewness toward the lower frequency range. In addition, most of the features exhibited the higher level of variability in PD patients compared with the healthy group.