Brain Network Activation Analysis Utilizing Spatiotemporal Features for Event Related Potentials Classification.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Event-related potentials (ERPs) are crucial for understanding brain activity.
• Automated classification of ERPs is challenging but essential for clinical applications.
• Existing methods may lack the spatiotemporal precision needed for accurate subgroup classification.

Purpose of the Study:

• To introduce an improved tool for automated ERP classification using spatiotemporal parcellation and functional brain network activation (BNA).
• To evaluate the tool's performance using the auditory oddball ERP paradigm.
• To assess the reliability and classification utility of the BNA scores.

Main Methods:

• ERPs were decomposed into dynamic spatiotemporal events.
• A group-level spatiotemporal event set was generated through clustering.
• A spatiotemporal reference BNA model was created from temporal relationships.
• Subject-specific scores were compared to the BNA model and used in a support vector machine classifier.

Main Results:

• BNA scores demonstrated good test-retest repeatability (intraclass correlation 0.51-0.82) for N100, P200, and P300 components.
• High classification accuracy was achieved when validating trained data on the same subjects (AUCs 0.93-0.95).
• Classification accuracy remained high for external validation groups from different centers (AUCs 0.81-0.85).

Conclusions:

• The improved spatiotemporal BNA analysis offers high classification accuracy for ERPs.
• The BNA method shows promise for diagnosing, monitoring, and developing drugs for neurological conditions.
• This tool can enhance the objective analysis of brain function in clinical settings.