EEG Connectivity and Power Spectral Density-Based Classification of Alzheimer’s Disease and Frontotemporal Dementia using 3D Convolutional Neural Networks

Accurate classification of Alzheimer’s Disease (AD) and Frontotemporal Dementia (FTD) from healthy controls (HC) is critical for early diagnosis and intervention. In this study, we explore two distinct approaches for extracting features from electroencephalography (EEG) signals for differentiating AD, FTD, and HC subjects using a 3D Convolutional Neural Networks. The first approach focuses on connectivity analysis to capture inter-channel interactions across different brain regions. When calculating connectivity between channels, we used alpha, beta, theta, delta band signals and also signal containing all bands. The second approach employs Power Spectral Density (PSD) analysis to quantify spectral power in EEG signal at different frequencies. Both methods independently construct 3D feature representations utilizing temporal slicing and concatenation, which are then used to train 3D CNN classifier. First healthy samples are classified from the samples contain Alzheimer’s Disease and Frontotemporal Dementia. Then classification was conducted for classifying between Alzheimer’s Disease and Frontotemporal Dementia. For classifying HC from AD and FTD the best performance got by the feature extraction of PSD values with 97.30% accuracy. And this approach also performed better for classifying between AD and FTD with accuracy of 97.00%. Throughout the process, we identified that the Alpha and Theta bands can serve as biomarkers for distinguishing Alzheimer’s Disease (AD) and Frontotemporal Dementia (FTD) from healthy controls (HC), while the Delta band is effective for classifying between AD and FTD.