Parkinson’s disease classification using gait characteristics and wavelet-based feature extraction

This paper proposes a method to classify idiopathic PD patients and healthy controls using both the gait characteristics of idiopathic PD patients and wavelet-based feature extraction. Using the characteristics of idiopathic PD patients who shuffle their feet while they are walking, we implemented the following three preprocessing methods: (i) we used the difference between two signals that each represented the sum of eight sensor outputs from one foot; (ii) we used the difference between the maximum and minimum records among the vertical ground reaction force outputs from the eight sensors under the left foot; and (iii) we used method (i) again, but on the signals each obtained from one foot by method (ii). After thus conducting the three preprocessing tasks, we created approximation coefficients and detail coefficients using wavelet transforms (WTs). Then, we extracted 40 features from these coefficients by using statistical approaches, including frequency distributions and their variabilities. Using the 40 features as inputs to the neural network with weighted fuzzy membership functions (NEWFM), we compared the performances of the three abovementioned methods. When idiopathic PD patients and healthy controls were classified using the NEWFM, the accuracy, specificity, and sensitivity of the results were, respectively, as follows: 74.32%, 81.63%, and 73.77% by method (i); 75.18%, 74.67%, and 75.24% by method (ii); or 77.33%, 65.48%, and 81.10% by method (iii).