A Model for the Detection of Abnormal Regions in Quantitative Cerebral Maps with Application to Myelin Water Fraction Maps

• Activity Number: 215
• Type: Invited
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Sandra Milena Hurtado Rúa*+
• Companies: Weill Medical College of Cornell University
• Keywords: Bayesian statistics ; Classification problems ; MRI ; T2-relaxometry ; MCMC ; Brain imaging

Abstract

T2-relaxometry is an increasingly popular MRI technique that can separate the contribution of various tissue components in the brain, i.e; it can quantify myelin water fraction (MWF) maps. These maps have wide applicability in neurological disorders like multiple sclerosis (MS). Clinically, a group of experts classify lesions after a visual analysis of individual maps. Statistical models have to potential to classify lesions on white and gray matter brain regions in a multi-subject design. Statistical models that identify abnormal regions while account for the spatial distribution of the brain maybe beneficial for diagnostic and clinical care. This talk proposes a Bayesian hierarchical model for the analysis of quantitative cerebral maps. A generalized linear classifier is proposed accounting for spatial ROI variability on normalized brain maps (fixed number of ROI per image). A flexible structure is considered by including symmetric and asymmetric link functions. A simulation studies the model performance. The model has potential application in multiple sclerosis research.