A Comprehensive Review of Parkinson's Disease Recognition from EEG Using Attention-Based Sparse Graph Convolutional Neural Networks

Abstract

Parkinson’s Disease (PD) is a progressive neurodegenerative disorder characterized by motor and non-motor impairments, requiring early and accurate diagnosis for effective management. Electroencephalography (EEG) has emerged as a non-invasive and cost-effective tool for detecting neural abnormalities associated with PD. However, conventional machine learning approaches often fail to capture the complex spatial-temporal relationships in EEG signals. Recently, attention-based sparse graph convolutional neural networks (ASGCNNs) have demonstrated significant promise in improving classification performance by modeling functional brain connectivity and selectively emphasizing informative EEG channels. This review provides a comprehensive analysis of state-of-the-art methodologies (2020–2023) for PD recognition using EEG, focusing on attention mechanisms and graph-based deep learning architectures. It examines advances in feature extraction, graph construction strategies, and interpretability. Comparative analysis highlights that ASGCNN models outperform traditional CNN and RNN approaches by leveraging channel interdependencies and sparsity constraints. Furthermore, this study identifies key challenges such as dataset limitations, model generalization, and clinical applicability. Future directions include hybrid architectures integrating transformers, multi-domain feature fusion, and explainable AI. The findings suggest that attention-based GCN frameworks represent a transformative approach for accurate, scalable, and interpretable EEG-based PD diagnosis.