Dynamic reconstruction of electroencephalogram data using RBF neural networks

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Electroencephalography (EEG) signals are complex and nonlinear, posing challenges for traditional analysis.
• Machine learning offers a promising avenue for overcoming these limitations in EEG analysis.

Purpose of the Study:

• To develop a novel approach using Radial Basis Function (RBF) neural networks optimized by Particle Swarm Optimization (PSO) for EEG signal reconstruction.
• To extract age-related neural characteristics from reconstructed EEG dynamics.
• To identify fixed points in the reconstructed neural system for quantitative analysis.

Main Methods:

• Collected EEG data from 142 participants across various age groups.
• Preprocessed EEG signals using bandpass filtering (1-35 Hz) and Independent Component Analysis (ICA).
• Trained RBF neural networks on EEG time-series data, optimizing parameters with PSO to identify fixed points.

Main Results:

• Achieved high accuracy in EEG signal reconstruction with a normalized root mean square error (NRMSE) of 0.0671 ± 0.0074.
• Confirmed the model's ability to capture oscillations through spectral and time-frequency analyses.
• Identified distinct age-related patterns in the fixed-point coordinates of the RBF network.

Conclusions:

• Fixed-point coordinates derived from RBF networks serve as quantitative markers for aging.
• The proposed method provides insights into age-dependent changes in brain dynamics.
• This computationally efficient approach has potential applications in neurological diagnosis and cognitive research.