EEGAlzheimer’sNet: Development of transformer-based attention long short term memory network for detecting Alzheimer disease using EEG signal

Abstract

A previous diagnosis of Alzheimer's disease (AD) in its initial stages is needed for patient care because it helps patients adopt preventative measures before irreversible brain damage occurs. Several studies have used computers to detect AD, although hereditary results limit most computer detection methods. There is no straightforward method to screen for AD, partly because the condition is difficult to diagnose and sometimes requires costly and occasionally intrusive testing that is uncommon outside of highly specialized clinical settings. Therefore, this study implements a deep learning strategy for detecting AD with the help of the “Electroencephalogram (EEG)” signal. Initially, the required EEG signal is obtained from traditional online databases and then applied to the 3-level “Lifting Wavelet Transform (LWT)” decomposition to decompose the signal into many wavelets. From the decomposed signal, the temporal features are retrieved by a “Recurrent Neural Network (RNN)”, and the spatial features are extracted from a “Multi-scale dilated Convolutional Neural Network (CNN)”. Further, the Enhanced Wild Geese Lemurs Optimizer (EWGLO) algorithm is implemented to find the optimal weight value for acquiring the weighted stacked features. These resultant weighted stacked features are applied to the semi-detection stage, where the “Optimized Transformer-based Attention Long Short Term Memory (OTA-LSTM)” model is utilized to detect AD. In the detection stage, parameter optimization takes place to increase the performance of the detection process using the same EWGLO. The designed model is validated with various performance metrics to show the effective outcome. Moreover, the developed model attains 96% and 98% in terms of accuracy and MCC. Throughout the validation, the offered model shows enriched performance when compared with other-state-of-art methods.