DIFFERENTIAL DIAGNOSIS OF PATIENTS WITH NEURODEGENERATIVE DISEASES AND HEALTHY SUBJECTS BY ANALYZING STRIDE TIME INTERVALS

Neurodegenerative disease (ND) symptoms are often associated with motor impairments. Analyzing gait fluctuations as a non-invasive and cost-effective approach is advantageous to distinguish healthy subjects from patients with ND. Previous studies have used time series information of gaits, including stride, stance, and swing intervals of both legs, and the utmost focus was on binary classification. However, this study selected only the right stride time intervals of patients and healthy subjects and proposed a computer-aided diagnostic system for two-class and multi-class classification. According to the type of classification, statistical features were extracted from the time domain, time-frequency domain using 2-level discrete wavelet transform (DWT), or both of them. The discriminative features were fed into the K-nearest neighbor (KNN) classifier. This method was evaluated on a publicly available dataset containing gait signals from healthy control subjects (CO) as well as patients with amyotrophic lateral sclerosis (ALS), Parkinson’s disease (PD), and Huntington’s disease (HD). It achieved an outstanding mean classification accuracy between 92% and 98% for binary classes. For the four-class classification, two approaches were employed. First, three groups (CO vs. ALS vs. (PD, HD)) were classified with an accuracy of 79.37%. In the second step, PD and HD were distinguished with 82.8% accuracy using a leave-one-out cross-validation approach. The results confirm the effectiveness of our proposed method at detecting ND patients in the early stages of the disease.