A0933
Title: Bayesian mixed-effects models for repeated EEG time-frequency data
Authors: Hyung Park - New York University School of Medicine (United States) [presenting]
Thaddeus Tarpey - New York University (United States)
Xiaomeng Ju - New York University (United States)
Abstract: Standard machine learning approaches in neuroimaging often treat functional data such as EEG as unstructured arrays, neglecting the hierarchical and repeated-measurement structures common in neuroscience studies. This is particularly limiting in cognitive and pain research, where EEG signals are collected across multiple conditions within subjects. A Bayesian mixed-effects framework is proposed for analyzing time-frequency representations of EEG data, which explicitly models both between-subject and within-subject variability for these two-way functional data. The approach incorporates covariate adjustment (e.g., age, sex, diagnosis), accounts for complex experimental designs, and provides full posterior uncertainty quantification. By borrowing strength across repeated measures and imposing structure through hierarchical modeling, this framework improves both inference and prediction compared to unstructured methods. Applications to EEG-based studies of alcoholism and pain illustrate the utility of this approach.
