Contrast Dyes and Cancer Maps: Approaches to Image Registration with Appearance Changes in Breast CT Scans

Abstract

Image registration has taken a prominent stance in the computer science research community because of its applications in navigation, augmented reality, and medicine. We study image registration in two sets of CT scans of the human breast, one before and one after the introduction of a contrasting dye designed to highlight mobile, or tumorous, cellular regions. The images are taken using a dedicated breast CT scanner. This thesis is the narrative behind the most effective models as they apply to contrast-enhanced breast image pairs. We focus extensively on regularization for appearance changes and experiment with affine, B-spline, diffeomorphic, and neural network-based registration.