Image registration illustrative results

In many medical imaging problems there is a need to align one image with another. For example if you acquire one X-ray CT scan of a patient who is holding their breath after exhaling, and another scan when they hold their breath at inhale, then you can register these two images to each other to help quantify motion that occurs during respiration (breathing). Understanding this motion can help in applications like radiation therapy for treating patients with lung cancer. Prof. Fessler and his graduate students are developing improved image registration tools using techniques from the field of image processing (interpolation, models), from statistics (developing measures of image similarity) and from optimization theory. This research is supported by the NIH National Cancer Institute and is performed in collaboration with scientists in the UM Department of Radiology and the UM Department of Radiation Oncology.

One challenge in nonrigid image registration is constraining the estimated deformation to preserve topology. Conventional 3D image registration using B-splines leads to mathematical models where tissues "fold" on themselves, which are physically implausible deformations. Prof. Fessler's graduate student Se Young Chun has developed an improved image registration algorithm (presented at ISBI 2008) that constrains the deformation to be locally invertible by including a special regularization function in the cost function that is optimized. His method is faster than the Jacobian penalty functions used previously and yields improved image registration results that are much more realistic for breathing motion.

Here are some links to page(s) that illustrate image registration. Some tools for image registration are available in the Tomography toolbox.

Back to Fessler home page