Integrating Advanced EEG Features with Graph Convolution Transformers for Alzheimer’s Detection

Alzheimer’s Disease (AD) poses significant challenges for early diagnosis due to its complex nature. Early detection is vital for improving patient outcomes. This study introduces an innovative approach for AD detection using graph convolution transformer networks (GCTNs), combining Graph Convolutional Networks (GCNs) and Transformer-based Convolutional Networks (Transformers) to analyse electroencephalography (EEG) data. Feature extraction includes standardizing EEG features, calculating statistical measures (mean, standard deviation, skewness, kurtosis), evaluating nonlinear characteristics (fractal dimension, sample entropy), and assessing frequency band ratios (theta-beta ratio, alpha-theta ratio). A mutual information-based adjacency matrix is constructed, representing the relationships between EEG features, and forming a graph where nodes are channels and edges indicate their mutual dependencies. Three models are assessed: a hybrid GCN-Transformer model (GCTN), a GCN-only model (GCN), and a Transformer convolution model (TN). The GCTN model integrates GCN and Transformer layers to leverage both local and global feature relationships, while the other models use GCN and Transformer layers individually. Model performance is evaluated based on classification accuracy, precision, recall, F1-score, and Receiver Operating Characteristic (ROC) curves with Area Under the Curve (AUC) values. The GCTN model attains an accuracy of 0.8073 after 3000 epochs, consistently outperforming the GCN and TN models in terms of precision, recall, and F1-scores. This model enhances classification accuracy for Alzheimer’s Disease detection. The study highlights the potential of integrating GCN and Transformer architectures to improve AD detection, offering promising prospects for clinical diagnostics and future research on neurodegenerative diseases.