Wavelet-Based Convolutional Neural Network for Parkinson's Disease Detection in Resting-State Electroencephalography

Electroencephalography (EEG) is not commonly used for Parkinson's Disease (PD) detection and diagnosis. However, it has been recently indicated in the literature that EEG may present unique biomarkers and features of the disease. In the current study, we introduce a Convolutional Neural Network (CNN) framework that exploits the wavelet domain of resting-state EEG in order to classify subjects into PD and Healthy Controls (HC). It was observed that PD exhibits a continuous uniform fading of the low wavelet scales as compared with HC. In addition, the proposed CNN approach was able to detect PD with a 4-fold as well as 10-fold cross-validation performance of up to 99.9% surpassing the-state-of-the-art deep learning-based architectures.