Characterization and biomechanical analysis of pulmonary disease using varifold-based CT image registration

Radiation therapy (RT) is one of the most common treatments for lung cancer. Image registration may be used to update the radiation treatment plan by deforming the original RT plan to accommodate for the anatomical shape changes. Traditional registration algorithms can not handle large anatomical changes well, and often some pre-knowledge is required. This research studied how large deformation varifold-based image registration could be adapted to register lung images. We call our image registration approach the pulmonary vessel and surface varifold-based registration algorithm (PVSV). We found that PVSV could handle lung shapes with substantial differences and is robust to missing information. The PVSV method achieved the best performance compared with previous image registration methods applied to lung images with a higher success rate.

The second focus of this dissertation was to determine which image registration algorithm performed best for registering pulmonary CT scans of patients with chronic obstructive pulmonary disease (COPD). In our work, we used data from SPIROMICS project. We compared and contrasted the registration algorithms performance of four state-of-the-art image registration algorithms. Biomechanical features were extracted from the transformations. Results show statistically significant increasing or decreasing trends for some of the features as a function of GOLD stage globally and on a lobe-by-lobe basis. These trends are consistent for all registration algorithms suggesting the robustness of biomechanical properties extracted by image registration and the authenticity of the detected trends.