Detection of mild cognitive impairment from electroencephalography (EEG) signals using discrete wavelet transformation leaders, spectral analysis and deep learning model

Mild Cognitive Impairment (MCI) is a neuropsychological syndrome characterized by cognitive disorders. Early diagnosis of MCI is important as it can slow the progression of the disease to Alzheimer's disease. Accurate and early diagnosis of the disease requires extensive testing and long-term observations. These can be subjective, expensive, or incomplete. EEG is a strong candidate in the diagnosis of diseases with its advantages such as being based on findings, non-invasive, and less costly. In this study, a new EEG-based model is proposed that can effectively detect MCI with higher accuracy by using deep learning, machine learning, and different feature extraction methods. In the study, EEG signals were denoised using Multiscale Principal Component Analysis (MSPCA), and Data Augmentation (DA) was performed to increase the size of the dataset. Then, Discrete Wavelet Transform Leaders (DWT leaders), Periodogram, Welch, and Multitaper feature extraction methods were used to extract features from the EEG signals. Validation of the model was carried out with the 10-fold cross-validation method. Finally, the performances of Bi-directional Long Short-Term Memory (BiLSTM), Decision Tree (DT), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN) algorithms are compared. The model, which integrates the multitaper spectral analysis and the BiLSTM deep learning algorithm, showed the highest performance as a result of the experiments. The proposed model achieved promising performance with 98.97% accuracy, 98.34% sensitivity, 99.67% specificity, 99.70% precision, 99.02% f1 score, and 97.94% Matthews Correlation Coefficient (MCC). Keywords: Deep learning, DWT leaders, Electroencephalography (EEG), Mild Cognitive Impairment (MCI), Power Spectral Analysis.