Exploiting the Differential Wavelet Domain of Resting-State EEG Using a Deep-CNN for Screening Parkinson's Disease

In this paper, a deep Convolutional Neural Network (CNN) of 20 layers was deployed to exploit the features extracted from the Laplacian of the Wavelet transform of a resting-state EEG time-series dataset in order to distinguish persons with Parkinson's disease (PD) from healthy controls (HC). It was shown that PD presents significant discriminative changes in the differential Wavelet domain as compared to HC especially at intermediate and lower scales. Due to the observed discrepancies, the deep CNN method was able to identify subjects with PD at a 4-fold as well as 10-fold cross-validation accuracy, sensitivity, and specificity of up to 99.9% surpassing most of the-state-of-the-art deep learning-based architectures.