Classifying Parkinson's Disease Using Resting State Electroencephalogram Signals and UEN-PDNet

• Published in: IEEE access 2023, Vol.11, p.107703-107724
• Main Authors: Rizvi, Syed Qasim Afser, Wang, Guojun, Khan, Asif, Hasan, Mohammad Kamrul, Ghazal, Taher M., Khan, Atta Ur Rehman
• Format: Article
• Language: English
• Subjects: Brain modeling; CNN; Computer architecture; Convolutional neural networks; Deep learning; EEG; Electroencephalography; Feature extraction; Neural engineering; Neuroimaging; Parkinson's disease
• ISSN: 2169-3536
• DOI: 10.1109/ACCESS.2023.3319248

Parkinson's Disease (PD) in the set of neuro-degenerative disorders stimulates due to the loss of dopaminergic neurons from the substantia nigra. Electroencephalogram (EEG) signals are being extensively utilized for diagnosing PD. The existing approaches extract the features using various frequency transformations that lose valuable signal information. An optimized Deep Convolutional Neural Network (CNN) inspired by the encoder part of U-Net architecture is proposed for classifying PD incorporating the resting electroencephalogram (EEG) signal dataset. The proposed model follows the U-Net architecture for extracting the features from the signals. The EEG recordings are taken from two datasets: the University of Mexico (UNM) EEGs and the University of California San Diego (UCSD) resting state dataset. The EEGs are pre-processed with a basic pre-processing pipeline, then separated into single channels, plotting each channel as a simple graph. These graphs are then fed to the proposed 23-layered convolutional neural network (CNN) for classifying PD from the normal control. Consequently, the model achieved maximum values of 93.10%, 93.18%, 93.09%, and 0.9313 of accuracy, precision, recall, and F1-score respectively for the UNM dataset, whereas, 97.90%, 98%, 97.87% and 0.9794 of accuracy, precision, recall, and F1-score respectively for UCSD dataset. The results show improved scores compared to the individual Machine Learning and CNN models applied on the same datasets.