Robust and Gaussian spatial functional regression models for analysis of event-related potentials

Area of Science:

• Neuroscience
• Biostatistics
• Statistical modeling

Background:

• Event-related potentials (ERPs) are crucial for understanding brain responses to stimuli.
• Traditional ERP analysis methods often overlook complex correlations across electrodes and subjects, potentially leading to missed insights and sensitivity to outliers.
• Existing methods may not fully capture the intricate spatiotemporal dynamics of electrophysiological data.

Purpose of the Study:

• To develop a novel Bayesian spatial functional regression framework for analyzing entire ERPs.
• To simultaneously account for multilevel correlation structures (spatial, temporal, and between subjects) in ERP data.
• To provide robust and adaptive statistical inference for ERP analysis, including outlier detection and model selection.

Main Methods:

• A Bayesian spatial functional regression framework using mixed models to analyze ERPs.
• Incorporation of basis-space Matérn assumptions for spatial correlation and correlated normal-exponential-gamma (CNEG) priors for regularization.
• Development of Gaussian and robust (heavy-tailed) models with predictive methods for model selection.
• Implementation of global tests and multiplicity-adjusted pointwise inference for spatiotemporal regions.

Main Results:

• The proposed framework successfully models entire ERPs as spatially correlated functional responses.
• It effectively accounts for multilevel correlation structures, including spatial and temporal dependencies.
• Analysis of smoking cessation data revealed significant effects across different visual stimuli types, highlighting the framework's utility.

Conclusions:

• The novel Bayesian framework offers a powerful and robust approach to analyzing complex ERP data.
• It overcomes limitations of traditional methods by accounting for spatiotemporal correlations and outliers.
• This approach provides comprehensive inference, from global effects to specific waveform components, with applications in neuroscience and clinical studies.