ADNET: A 1D-CNN Feature Fusion-based Method for Alzheimer’s Disease Detection Using EEG Signals

Alzheimer’s disease (AD) is a progressive neurodegenerative disease that significantly impairs cognitive abilities. There is no cure for AD except for early detection and effective intervention. Electroencephalography (EEG) is a non-invasive and relatively inexpensive method to diagnose AD at an early stage; however, automated detection of AD using EEG signals remains a challenge due to noise incurred during data acquisition, extraction of features with high variance between classes, and accurate classification. We present a customized architecture Alzheimer’s Detection Net (ADNET) based on one-dimensional convolutional neural network (1D-CNN) for high interclass feature vector. The proposed method consists of three steps including preprocessing of EEG signals, feature extraction, and classification. In the first step, EEG signals are denoised using independent component analysis, Band-pass filter, and artifact subspace reconstruction. A customized 1D-CNN-based architecture ADNET is proposed for feature extraction. Both handcrafted and ADNET features are then concatenated to form a feature vector, which is then passed to CNN for classification between AD and frontotemporal dementia. We applied the proposed method on the publicly available OpenNeuro dataset and achieved an accuracy of 94.88%. The ADNET-based proposed method outperformed the existing state-of-the-art methods of AD detection.