A Comprehensive Review of Parkinson's Disease Recognition Via Heterogeneous Split Attention-Based EEG and Siamese Graph Convolutional Attention Network

Abstract

Parkinson’s Disease (PD) is a progressive neurodegenerative disorder characterized by motor and non-motor symptoms that significantly impact patients' quality of life. Early and accurate diagnosis remains a major clinical challenge due to subtle symptom onset and variability across patients. Recently, Artificial Intelligence (AI), particularly deep learning techniques, has demonstrated significant potential in improving PD diagnosis through non-invasive modalities such as electroencephalography (EEG). EEG signals capture brain activity and provide valuable insights into neural dysfunction associated with PD. Advanced architectures such as attention-based convolutional neural networks, graph neural networks (GNNs), and Siamese learning frameworks have enhanced classification performance by modelling spatial-temporal dependencies and inter-channel relationships. For instance, attention-based sparse graph convolutional neural networks have shown improved diagnostic accuracy by capturing functional connectivity among EEG channels. Additionally, hybrid CNN-LSTM models effectively extract both spatial and temporal features from EEG signals, further improving classification robustness. This paper presents a comprehensive review of recent advances in PD recognition using heterogeneous split-attention EEG models and Siamese graph convolutional attention networks. It highlights trends, challenges, and future research directions, emphasizing the role of explainable AI and multi-modal integration in enhancing diagnostic reliability.