Graph based novel features for detection of Alzheimer’s disease using EEG signals

• Published in: Biomedical signal processing and control Vol. 103; p. 107380
• Main Authors: Sharma, Ramnivas, Meena, Hemant Kumar
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.05.2025
• Subjects: Alzheimer’s disease (AD); Convolutional neural network (CNN); EEG signals; Feature extractions; Graph Fourier transform; Graph wavelet transform; Machine learning models; Graph Fourier transform; Graph wavelet transform; Alzheimer’s disease (AD); EEG signals; Convolutional neural network (CNN); Machine learning models; Feature extractions
• ISSN: 1746-8094
• DOI: 10.1016/j.bspc.2024.107380

Alzheimer’s disease (AD) is a progressive neurodegenerative condition of the brain, leading to neuronal death and dementia. Its global prevalence is on the rise, prompting extensive research into automated detection methods using brain signals. This research integrates graph-based features with traditional methods to increase the precision of EEG signal-based early Alzheimer’s disease (AD) detection. Conventional techniques frequently neglect the complex functional connections within the brain during different activities by analyzing EEG channels separately. To address these shortcomings, this study introduces an innovative method based on graph signal representations of EEG signals that take into account the interrelationship within the brain. It leverages graph spectral features, including the graph Fourier transform (GFT) and the graph wavelet transform (GWT), and incorporates statistical features such as mean, maximum, minimum, median, standard deviation, and kurtosis. Following preprocessing and feature extraction from the EEG signals of 24 healthy individuals, 24 AD patients (Dataset-I), and 12 healthy subjects, along with 80 AD patients (Dataset-II), different machine learning classifiers are applied. Comparative analysis reveals that random forest (RF) and convolutional neural networks (CNN) achieve the highest performance, with accuracies of 99.75% and 99.48% on the datasets, respectively, demonstrating their effectiveness in constructing an AD diagnostic system. This study combines graph-based and statistical features with the GWT method for EEG signal representation and the RF classifier for classification, with comparative accuracy in AD detection. These findings demonstrate how graph-based features may be used to enhance early disease detection and treatment. •Comparative analysis of Alzheimer’s disease (AD) classification using ML classifiers and deep learning.•Graph signal processing concepts are used to analyze EEG data for AD detection.•GFT and GWT graph-based features are evaluated for classification purposes.•Proposed graph signal features capture brain connectivity and temporal dynamics.•Graph-based features enhance AD detection accuracy from EEG signals.•RF and CNN classifiers achieved 99.75% and 99.48% accuracy, respectively.