Identification of Alzheimer's disease from central lobe EEG signals utilizing machine learning and residual neural network

Cognitive and behavioral deficits are some of the symptoms of Alzheimer's disease, a neurological disease caused by brain deterioration. Early diagnosis of the disease minimizes the disease's progression. In this way, patients' living standards can be maintained. An experienced specialist will assess the results of grueling diagnostic tests and a costly diagnosis process. This motivation led to the creation of a fully automated computer system, which uses the proposed methods to detect Alzheimer's disease from EEG signals acquired from a minimum number of electrodes: three central lobe electrodes. This work presents the advantages of implementing the proposed system. Two different datasets were presented to evaluate the system's performance. After preprocessing the raw EEG data, wavelet transforms were applied to calculate statistical properties. First, the paper discusses the performance of different traditional machine learning classifiers: Diagonal Discriminant Analysis, Support Vector Machine (LSVM, RBF), K-Nearest Neighbors, Random Forest, Logistic Regression, and Naïve Bayes. By comparing the classifiers’ performance, it is found that Naïve Bayes and LSVM classifiers yield the highest performance of 96.55% and 95.69% when applied to the first dataset and 96.55% and 94.52% for the second dataset, respectively. “ResNet-50” Convolutional Neural Network classifier was implemented to demonstrate the capability of constructing an effective AD diagnosing system. It yields an accuracy of 97.8261% when applied to the first dataset. This work shows how Deep Learning improves the performance of classification in early medical diagnosis, which will enhance different diseases’ treatment, as “Resnet – 50” convolutional neural network classifier outperformed Naïve Bayes Classifier.