Cascade Wavelet Transform and Optimal Modified Deep Convolutional Neural Network for Alzheimer’s Disease Detection in EEG

Abstract

Alzheimer's disease has become the most widespread neurological brain ailment and may be diagnosed using a variety of medical approaches. EEG signals are frequently used to diagnose cognitive issues, especially if there is a discrepancy in the diagnosis following the first clinical investigations. Nonetheless, there is proof that the EEG may accurately detect Alzheimer's illness. The EEG diagnosis of Alzheimer's disease (AD) is growing in frequency. A novel resting-state EEG classification method for Alzheimer's disease (AD), moderate cognitive impairment (MCI), and healthy control (HC) is presented in this work. Early diagnosis of mild but noticeable cognitive impairments that do not significantly affect day-to-day functioning may lower the death and morbidity rates associated with prodromal Alzheimer's condition. During such studies, disturbances and interference from the EEG dataset are eliminated using a band-pass elliptic digital filter (BEF). To get the properties of the EEG signal, the filtered data was separated into frequency bands using the Cascade wavelet transform (CWT) technique. Then, by including a variety of signal properties into feature vectors, the CWT technique was utilized to improve diagnostic performance. The improved EEG was categorized using an Optimal modified Deep Convolutional Neural Network (OMDCNN) once the necessary dataset was created. Here, the ideal MCNN hyperparameter is determined using the modified clonal selection algorithm (MCSA), which might increase the classifier's accuracy. Finally, the sensitiveness, accuracy, quality of diagnosis, and the area beneath the receiver operating characteristic (ROC) curve were additionally computed in order to compare and evaluate the performance of the various recommended approaches.