2017-04-07

Subhadip Pal, PhD, Department of Bioinformatics and Biostatistics, University of Louisville

"A Distributional ICA Model for Analyzing Multimodal Neuroimaging Data"

Research advances in neuroimaging and pathophysiology hold great promises to revolutionize diagnosis and treatment of mental health diseases. There is an emerging interest in conducting studies that focus on investigations of mental disorders leading to an expanded depth of multimodal brain imaging data. A popular approach to characterize brain functional networks in fMRI studies is independent component analysis, a powerful method to reconstruct latent source signals from their linear mixtures. However, the standard ICA framework is not directly applicable to multi-modal imaging data, especially when data generated from different imaging modalities have different scales (continuous/discrete), representations (scalar/array/matrix) and variability. This creates great difficulties for effective joint analysis to extract combined features across dimensions. In this project, we develop a novel approach, termed as Distributional Independent Component Analysis (DICA), for addressing the aforementioned issues to analyze multi-modal data. We perform statistical inference on the model parameters under the Bayesian framework and develop a posterior computation algorithm based Markov chain Monte Carlo (MCMC) method. We demonstrate the advantages of our methods via extensive simulation studies and analyses of an fMRI dataset and a DTI dataset.

Stay connected TwitterFacebookLinkedInYouTubeInstagram