Image-Derived Forces in an Elastic Image Registration Model, 2021

Advisors: Erik Hanson, Erlend Hodneland

Short description of the project:

In this work, we explore the role of image-derived forces in an elastic image registration model, and investigate the possibility of accurately estimating displacement fields. Further, we propose a novel, physically motivated, image registration method where the resulting displacement is controlled by the boundary conditions. This is in contrast to traditional methods, where the displacement is driven by an unphysical image-derived body force. Several registration experiments are done, both to highlight the role of image-derived forces, and to demonstrate the capabilities of the novel method. Additionally, the method is used to explore the effects of optimizing over tissue parameters. Experiments show us that there is a poor agreement between similarity error and displacement field error when performing traditional elastic registration. The novel method produces satisfactory visual registration results, comparable to existing methods. The method ensures that the resulting displacement field is physically possible and obeys the governing equations, however, this is at the cost of a worse performance in terms of minimization of the chosen distance measure.

Link to the thesis.