Athinoula A. Martinos Center for
Biomedical Imaging,
Massachusetts General Hospital,
Harvard Medical School
ABSTRACT
The last two decades have seen tremendous advances in computational techniques for automatically segmenting (i.e., delineating structures in) magnetic resonance (MR) images of the brain. However, very little of these methodological developments has found its way into clinical practice. This is because computing reliable segmentations for individual patients in clinical routine is very different from analyzing groups of subjects in well-controlled scientific studies. Clinical images are based on other MR contrast mechanisms; are not standardized across imaging sites; and often contain large pathologies that violate the computational models that are used. Many of the leading brain segmentation techniques are also much too slow to be useful in a clinical context. In this talk I will discuss ongoing research in my group on a Bayesian segmentation approach in which computational models of brain anatomy and lesion shape are combined with independent, adaptive models of the imaging process. I will show that this approach holds the potential to yield an extremely fast and widely applicable clinical tool; specific applications in multiple sclerosis and radiation therapy of brain tumors will be highlighted.
Host: Stanislav Emelianov, Ph.D.
Tuesday, Feb. 7
10:30 a.m.
McIntire Room 3115,
Whitaker Bldg
Videoconference:
Emory: HSRB E182
Georgia Tech: TEP 104