Invited Talk by Kayhan Batmanghelich

We present a novel approach to joint analysis of heterogeneous medical data, including images and genetic markers. The emerging field of imaging genetics aims to 1) use images as intermediate data to uncover underlying biological mechanisms of disease; 2) use known genetic markers of the disease to discover associated properties of medical images (so-called phenotype discovery). In this talk, we mainly focus on the first aim, namely using imaging data as an intermediate representation of the disease. Using the language of probabilistic graphical models, we formulate a unified framework that exploits the rich information in the MR images of the brain to detect genetic variants that are implicitly related to the Alzheimer's disease. We demonstrate how imaging can help identify relevant genetic variants that existing methods fail to detect, and illustrate advantages of the joint modeling over methods that characterize imaging and genetic associations separately. We briefly discuss the second goal of imaging genetics in the context of the Chronic Obstructive Pulmonary Disease (COPD) where genetic information is used to discover commonly occurring imaging patterns in the lung CT images of the patients.