Seminars/Colloquia/Invited Talks

Seminars

James Gee

Shape-optimizing Diffeomorphisms for Computational Anatomy

PLACE:	Clark 110
EVENT:	CIS Seminar Series
DATE:	September 28, 2005
TIME:	3:00 - 3:45

Abstract

We propose Diffeomorphometry (DM), morphometry with diffeomorphisms. Diffeomorphometry is a rigorous mathematical framework for statistical analysis of shape and shape coordinate systems derived from images. Diffeomorphometry is characterized by computing diffeomorphisms and their inverses in the Lagrangian frame, representing diffeomorphisms by their initial conditions, and combining initial conditions to evaluate the statistics and neighborhood relationships of shapes. Statistical simulations based on these statistics are also derived.
The diffeomorphism group, G, has several benefits in medical image registration. Foremost, diffeomorphisms may be used to optimize variational energies, including landmark and image similarity measures, while maintaining domain topology. Distance measurements between images and between diffeomorphisms are symmetric. The third important advantage is that diffeomorphic transformations are composable. Our work on Diffeomorphometry develops:
1. Intrinsically symmetric image registration and geodesic interpolation;
2. Optimal shape population atlases with a strong notion of the average coordinate system;
3. Well-defined variational principles for performing shape optimization and morphometry studies with diffeomorphisms;
4. A statistical framework for diffeomorphisms allowing simulation of diffeomorphic shape change.
These contributions are linked by a computational anatomy framework and exploit the composability and metric properties of diffeomorphisms. The methods use both image and expert landmark similarity measures. We apply the methods to studying shape changes between different species, interpolating anatomy in time, to study the temporal changes caused by neuroanatomical disease and to optimize the volumetric shape of histological sections.

Brief biography

James Gee, Ph.D., is Associate Professor of Radiologic Science in the Department of Radiology at the University of Pennsylvania School of Medicine. He received the B.S. degrees in Computer Science and Electrical Engineering from the University of Washington, Seattle, and the M.S. degree in Electrical Engineering from the same institution. He holds a Ph.D. in Computer and Information Science from the University of Pennsylvania, where he was awarded an R01 grant on medical image analysis while still pursuing graduate studies. Best known for contributions to computational anatomy, his research interests include biomedical imaging, probabilistic and geometric modeling, pattern analysis, and scientific computing.