Predicting Age From Brain EEG Signals-A Machine Learning Approach.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Brain age gap estimate (BrainAGE) typically uses MRI, with limited exploration of electroencephalography (EEG) and machine learning (ML).
• Investigating age-related changes in EEG signals is crucial for understanding brain aging.
• Predicting chronological age and BrainAGE using EEG and ML requires advanced feature extraction and modeling.

Purpose of the Study:

• To determine if age-related changes in brain EEG signals can be used for chronological age prediction.
• To develop and validate a machine learning framework for estimating BrainAGE from EEG features.
• To explore novel and extensive EEG feature extraction techniques for enhanced age prediction.

Main Methods:

• Utilized EEG data from 468 participants in the Tulsa-1000 longitudinal study.
• Employed five sets of preprocessed EEG features across channels and frequency bands.
• Applied a nested cross-validation (NCV) approach with stack-ensemble learning for age prediction.

Main Results:

• The stack-ensemble model achieved R² = 0.37 and Mean Absolute Error (MAE) = 6.87 years.
• A correlation of r = 0.6 was observed between chronological and predicted age.
• Feature importance analysis indicated widespread age predictors across various EEG feature types.

Conclusions:

• A rigorous ML framework with extensive EEG features enables reliable estimation of chronological age and BrainAGE.
• The developed methodology offers a robust alternative to traditional MRI-based brain age estimation.
• This framework can be extended to investigate EEG associations with other physiological responses and conditions.