Recent Advances in Parkinson's Disease Recognition Via Heterogeneous Split Attention-Based EEG and Siamese Graph Convolutional Attention Network: A Systematic Review

Abstract

Parkinson’s Disease (PD) is a progressive neurodegenerative disorder that significantly affects motor and cognitive functions. Early and accurate diagnosis remains a major challenge due to subtle symptom onset and variability across patients. In recent years, Electroencephalography (EEG)-based analysis combined with Artificial Intelligence (AI) techniques has emerged as a promising approach for non-invasive and cost-effective PD detection. Deep learning models, particularly Graph Convolutional Networks (GCNs), attention mechanisms, and Siamese architectures, have demonstrated strong capabilities in capturing complex spatial-temporal dependencies in EEG signals. This review focuses on recent advances in heterogeneous split attention-based EEG analysis and Siamese graph convolutional attention networks for PD recognition. These approaches effectively model brain connectivity and channel relationships, improving classification accuracy. For instance, attention-based sparse graph convolutional networks have been shown to enhance feature learning by focusing on significant EEG channels and functional connectivity patterns. Furthermore, hybrid models integrating CNN, recurrent networks, and attention mechanisms have improved robustness and generalization in PD detection tasks. Despite these advancements, challenges such as limited datasets, interpretability, and computational complexity persist. This paper provides a comprehensive review of recent methodologies, highlights key trends, and outlines future research directions for developing reliable and scalable PD diagnostic systems.