|
Abstract:
|
Advances in MRI technology now offer brain images of different modalities that capture brain organization from both functional and structural perspectives. Analysis of multimodal images involve some common goals. However, since these modalities have vastly different data characteristics, current analysis are usually performed using distinct analytical tools that are only suitable for a specific modality. In this talk, we present a novel Distributional Independent Component Analysis (D-ICA) method that provides a unified framework for decomposing multimodal imaging such as functional MRI (fMRI) and diffusion tensor imaging (DTI). Having a nice connection with the classic ICA, D-ICA represents a fundamentally new approach by performing ICA on the distribution level rather than on the raw imaging measurements. The D-ICA provides a general framework to extract neural features across imaging modalities that have different scales, representations, signal-to-noise ratios, and intensity. We develop a Bayesian approach and also a computational efficient approach for model estimation. We illustrate the method using extensive simulation studies and real-world multimodal imaging applications.
|
