*** BME Faculty Candidate ***
Ipek Oguz, Ph.D.*
Department of Radiology
University of Pennsylvania
Graph-based Segmentation for Medical Image Analysis
Image segmentation is one of the core problems in medical image analysis. Graph-based segmentation methods are attractive because of their computational efficiency as well as their guarantee to obtain the globally optimal solution of the cost function under certain conditions. Surface-based graph formulation further allows introducing a shape prior, which can be crucial to performance in many segmentation tasks. In this talk, I will present two recent graph-based segmentation algorithms for neuroimaging applications. The first is an algorithm for automated reconstruction of the cortical surface from MRI data, showing that graph-based segmentation is a significantly more accurate and significantly faster tool than FreeSurfer for cortical thickness studies. The second algorithm is focused on the segmentation of subcortical structures. The size and shape of these structures are used to derive important imaging-based markers in many neurological and psychiatric conditions. However, the large variability in deep gray matter appearance makes their automated segmentation from MRI scans a challenging task. This algorithm illustrates how machine learning techniques can be used in combination with graph-based methods for improved segmentation accuracy.
Faculty Host: Erin Buckley, Ph.D.