Difference between eyes-open and eyes-closed resting state quantitative electroencephalography (qEEG) for predicting cognitive impairment using deep learning.

Objective: Electroencephalography (EEG) is a method that offers detailed observations of electrical activities occurring in the brain's cerebral cortex. The EEG-derived brain signals can serve as a neurophysiological indicator for the early detection of dementia through quantitative EEG (qEEG) analysis. This study introduces a deep learning (DL)-based classification approach trained using the difference between qEEG time-frequency (TF) images of eyes-open resting (EOR) and eyes-closed resting (ECR) state EEG to detect mild cognitive impairment (MCI) and Alzheimer's disease (AD).

Patients and methods: The dataset comprised 16,910 TF images from 890 subjects, including 269 normal controls (NC), 356 with MCI, and 265 diagnosed with AD. Artifact-free EEG signals were converted into qEEG TF images using the Fast Fourier Transform (FFT), which captured various event-related alterations within the five (delta, theta, alpha, beta, gamma) EEG frequency sub-bands. EEG data was preprocessed using the EEGlab toolbox version 2022 within the MATLAB R2024a software environment. The preprocessed TF images, along with numerical age data, were used as input features in convolutional neural network (CNN) within a DL framework for classification.

Results: The performance metrics of the trained models for EOR - ECR in differentiating NC versus MCI, NC versus AD, and NC versus cognitive impairment (CI) (MCI + AD) classes were assessed using the test dataset from the subjects. The model NC versus CI yielded the best area under the curve (AUC), accuracy, sensitivity, and specificity with 0.95, 0.93, 0.97, and 0.92; NC versus AD was 0.88, 0.88, 0.89, and 0.86; and NC versus MCI was 0.85, 0.83, 0.9, and 0.81, respectively.

Conclusion: The findings suggest that using the difference between EOR and ECR qEEG states could be a practical approach to detect cognitive impairment in dementia research with DL. The trained models may serve as a supportive reference for clinicians in the future, not as a diagnostic tool, but as a decision-support system for early diagnosis of cognitive impairment.