Lifespan brain age prediction based on multiple EEG oscillatory features and sparse group lasso

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Neural dynamics change significantly throughout human development and aging.
• Electroencephalography (EEG) offers high temporal resolution for capturing neural oscillations, crucial for understanding these changes.
• Accurate brain age (BA) prediction from EEG is vital for identifying individuals at high risk within large populations.

Purpose of the Study:

• To develop an interpretable brain age prediction model using EEG.
• To clarify the lifespan representation of EEG oscillatory features (OSFs).
• To enhance practical clinical applications of BA prediction.

Main Methods:

• Extracted four groups of OSFs (aperiodic, periodic, power-ratio, relative power) from a multinational dataset (5-97 years).
• Mapped OSF trajectories and importance using GAMLSS and PCC.
• Extracted inter-oscillatory dependency coefficients (ODCs) via sparse group lasso.
• Fused OSFs and ODCs into a three-layer fully connected neural network (FCNN) for BA prediction, analyzing interpretability with Layerwise Relevance Propagation.

Main Results:

• The FCNN model incorporating ODCs significantly improved BA prediction accuracy (MAE = 2.95 years, R² = 0.86) compared to using OSFs alone (MAE = 3.44 years, R² = 0.84).
• Demonstrated the effectiveness of integrating OSFs and ODCs for more precise brain age estimation.

Conclusions:

• Proposed NEOBA, an interpretable brain age prediction model leveraging OSFs and ODCs.
• NEOBA shows promise for precise individual stratification through normative modeling in clinical settings.