CNN for a Regression Machine Learning Algorithm for Predicting Cognitive Impairment Using qEEG.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Electroencephalogram (EEG) signals provide insights into cerebral cortex electrical activity.
• EEG analysis, particularly quantitative EEG (qEEG), can serve as a neurophysiological biomarker for early dementia diagnosis.
• Mild cognitive impairment (MCI) and Alzheimer's disease (AD) are significant neurological disorders requiring early detection methods.

Purpose of the Study:

• To propose a machine learning methodology for detecting MCI and AD.
• To utilize quantitative EEG (qEEG) time-frequency (TF) images for cognitive impairment detection.
• To develop a classification model using eyes-closed resting state (ECR) EEG data.

Main Methods:

• A dataset of 16,910 TF images from 890 subjects (269 healthy controls (HC), 356 MCI, 265 AD) was used.
• EEG signals were transformed into TF images via Fast Fourier Transform (FFT).
• A convolutional neural network (CNN) processed image features, concatenated with age data for a feed-forward neural network (FNN) classifier.

Main Results:

• The HC vs MCI model achieved 83% accuracy, 93% sensitivity, and 73% specificity.
• The HC vs AD model achieved 81% accuracy, 80% sensitivity, and 83% specificity.
• The HC vs CASE (MCI + AD) model achieved 88% accuracy, 80% sensitivity, and 90% specificity.

Conclusions:

• The proposed machine learning models demonstrate potential as biomarkers for early detection of cognitive impairment.
• TF images derived from EEG signals, combined with age data, can effectively assist clinicians.
• This methodology offers a promising tool for early diagnosis in clinical settings.