Energy Functionals: Choices and Consequences For Medical Image Segmentation

August 18, 2011
Chris McIntosh | SFU

Medical imaging continues to permeate the practice of medicine, but automated yet accurate segmentation and labeling of anatomical structures continues to be a major obstacle to computerized medical image analysis. Though there are numerous approaches for medical image segmentation, one in particular has gained increasing popularity: energy minimization-based techniques, and the large set of methods encompassed therein. With these techniques an energy function must be chosen, segmentations must be initialized, weights for competing terms of the energy functional must be tuned, and the resulting functional minimized. There are a lot of choices involved, and their consequences are not always clear. In this talk I explore the different consequences of these choices, and provide novel methods that attempt to overcome two of the more significant problems encountered: local minima and parameter settings.

Speaker Details

Chris McIntosh received the B.Sc. (Honours) degree in Computing Science from Simon Fraser University (SFU), Canada in 2005. He is currently a Ph.D. candidate at the Medical Image Analysis Lab (MIAL) at SFU. Chris is the recipient of federal research scholarship awards from both the National Sciences and Engineering Research Council of Canada and the Canadian Institute for Health Research, as well as a provincial award from the Michael Smith Foundation for Health Research, which supported his research on medical image analysis since 2005, including co-authoring 3 book chapters (published by Wiley, Springer, and CRC) and 5 related conference papers on deformable models and deformable organisms focused on two specific applications: vessel and spinal cord segmentation (MICCAI 2006, CVPR 2006). His work on vascular segmentation and analysis was featured on the front page of the Vancouver Sun and other newspapers across Canada, and in the IEEE Intelligent Systems – In the News. Among his contributions to the field are the ITK Deformable Organisms, implementing the artificial-life framework for medical image analysis, with 5-star Insight Journal reviews and over 2400 downloads; as well as a method for visualization of high-dimensional medical data by optimally mapping it to a perceptually uniform color-space, with a corresponding co-authored paper in IEEE TMI. More recently he has been focusing on learning objective functions and optimization methods, resulting in a first-authored publication in IEEE TMI on building statistical deformable models using genetic algorithms, and other related conference papers (MCCAI 2007, ICCV 2011).