February 16th, 2012 @ 1130am Leo Grady (SCR-US)

Graph theory applied to medical image analytics

Radiologists and pathologists use geometrical information to make diagnostic and prognostic decisions about the anatomy they observe. As segmentation algorithms improve and datasets become larger, it becomes possible to learn image-based biomarkers with a precise and complete set of measurements. Traditional methods of quantifying shape emphasize classical measures of geometry, such as volume, surface area and curvature which may be evaluated with increasingly mature image segmentation methods. Additional information may be obtained by modeling biological structures as a network of pixels, which allows us to employ an arsenal of powerful network characterization tools to quantify structural and topological properties of biological structures. Finally, population studies to discriminate different subgroups are enabled by our ability to link different locations and attributes across individuals. In this talk I will present how we have been able to apply graph theory to generate robust and effective image segmentation methods, to produce correspondence between different individuals and how we can learn to discriminate population subtypes by going beyond geometrical features with advanced network characterization. Many of the applications shown will be drawn from neuroscience and oncology.

seminars/seminaritems/2012-02-16.txt · Last modified: 2012/02/13 12:30 by silberman