Skip to content. Skip to navigation
CIM Menus
 

A Bayesian Framework for 4-D Segmentation of Multiple Sclerosis Lesions in Serial MRI of the Brain


Colm Elliot
Department of Electrical and Computer Engineering McGill University

March 7, 2016 at  2:00 PM
McConnell Engineering Room 603

This thesis proposes methods for detection of the appearance, growth, and disappearance of pathology in serial Magnetic Resonance Images (MRIs) of the brain, specifically in the context of new T2-lesion formation and resolution in subjects with Multiple Sclerosis (MS). Detection of appearance and change in pathology in serial MRI is a difficult problem as information from multiple MRI images over multiple timepoints must be considered simultaneously, and MRI images are subject to image artifact that may not be consistent over serial scans of the same subject. The detection of change in pathology becomes a task of differentiating change of interest from apparent change arising from artifactual sources. Furthermore, new pathology may be small and subtle, and share intensity characteristics with healthy tissue. To robustly identify change of interest, this thesis develops two related Bayesian frameworks that consider multiple sources of information and uncertainty simultaneously: MRI intensities from multiple sequences, MRI intensity change over time, local spatial and temporal context, global anatomical context, as well a prior knowledge about the location and temporal dynamics of the pathology under investigation. The first is a pairwise method that considers consecutive timepoint pairs in turn, by detecting change in a follow-up timepoint with respect to a reference timepoint. The second is a unified framework that considers a full set of serial scans simultaneously to provide a 4-D spatio-temporal segmentation of individual lesions.