|
Abstract:
|
Mental health studies increasingly collect high-dimensional multimodal or longitudinal data to understand complex disease mechanisms or develop biomarkers. This leads to statistical challenges including preprocessing, dimension reduction, data integration, modeling/inference, computation, diagnostics and visualization. A representative instance is densely-sampled functional data, collected repeatedly over multiple patient visits or during intensive monitoring of which electroencephalography (EEG) is a typical example. EEG data from event-related potential (ERP) experiments have a complex, high-dimensional structure in which each of a long sequence of stimuli, or trials, generates an ERP waveform at multiple scalp locations. Traditional EEG analyses collapse the functional, longitudinal and spatial components by extracting key ERP features and averaging across trials and electrodes, resulting in significant information loss. Here I present ongoing work on a methodological framework which preserves more of the structural richness of such data and allows for a variety of inferential and interpretative tools. The techniques will be illustrated using studies of autism and schizophrenia.
|
