The Computer Science Colloquium

Robert Pless
(Washington University in St. Louis)

"Image Manifolds in Medical Imagery"

    Dynamic medical imagery often includes variations in images due to breathing and heartbeat. These cause image changes which, while complex, are often consistent over time; thus these images define a low dimensional manifold. Applying generic manifold learning techniques to real world images highlights several common shortcomings in current methods. First, using Euclidean distance (sum-of-squared pixel intensity difference) is usually a poor choice of image distance functions for natural images. Second, many natural image manifolds have a cyclic topology (and thus cannot be cleanly embedding into a Euclidean space). Third, natural data sets often include unlabeled examples from multiple, intersecting low-dimensional manifolds.

I will talk about several heuristic (and occasionally well founded) algorithms for choosing effective local image distance measures, finding minimal parameterizations for cyclic manifolds, and simultaneously clustering and parameterizing data from multiple intersecting manifolds. These have been brought together in an end-to-end application which automatically learns the 2D manifold structure of (ungated, free-breathing) cardiac MRI images of a patient, and uses the manifold structure of the images to regularize the image segmentation simultaneously in all frames.


Brief Biography:
Robert Pless is an Associate Professor of Computer Science and Engineering at Washington University in St. Louis. His research focus is the statistics and geometry of video, including anomaly detection and motion pattern analysis with applications to surveillance video and MR-imagery. Dr. Pless has Bachelors Degree in Computer Science from Cornell University in 1994 and a PhD from the University of Maryland, College Park in 2000. He chaired the IEEE Workshop on Omnidirectional Vision and Camera Networks (OMNIVIS) in 2003, and received the NSF CAREER award in 2006.

The Colloquium is supported by generous contributions from the Bloomberg, Information Builders, Inc., and Netlogic, Inc.