Constructing and Solving Variational Image Registration Problems

Successfully aligning medical images from a single modality or from multiple modalities enables subsequent processing, analysis, and/or visualization that can aid in the screening, diagnosis, prognosis, treatment, and monitoring of disease. In the past two decades, a vast amount of research has been performed to develop various models and computational techniques for image registration within a wide spectrum of applications. This presentation explores the approaches that have been developed for non-rigid registration using variational techniques. It focuses on three key aspects of the variational registration problem: dissimilarity measures, regularization, and rapid solution techniques. It presents new contributions in each area: in modifying dissimilarity measures to appropriately handle the changing regions of overlap between images, in combining two standard taxonomies of regularizers (homogeneous: diffusion, elastic, curvature; and non-homogeneous: image-driven and flow-driven), and in extending rapid solution algorithms that have been developed for specific regularizers (Fourier methods, Demons) to enable more general use.

Nathan Cahill spent 12 years as a research scientist at Eastman Kodak Company and Carestream Health, where he worked on image registration and fusion problems in consumer digital imaging, depth imaging, aerial imaging, and medical imaging. He is currently finishing his PhD in Engineering Science at the Institute for Biomedical Engineering at the University of Oxford, on the topic of variational image registration. In August, he will join the Center for Applied and Computational Mathematics at the Rochester Institute of Technology in Rochester, NY, as an associate professor. Nathan has nearly 20 conference and journal publications as well as 21 granted patents, and he is a Senior Member of IEEE.