Date of Award
5-14-2024
Degree Name
Doctor of Philosophy (PhD)
Degree Type
Dissertation
Abstract
The fundamental objective of radiation therapy is to deliver high doses of ionizing radiation to the target while minimizing the dose to surrounding normal tissues. Cone-beam computed tomography (CBCT) plays an essential role in ensuring the treatment precision by providing on-board volumetric images of the patient’s anatomy, facilitating target localization and patient positioning correction prior to treatment sessions. However, conventional CBCT imaging is susceptible to motion artifacts, potentially leading to inaccuracies in patient setup. To address this challenge, four-dimensional (4D) CBCT has been developed to capture time-resolved positional information of the patient, thereby reducing setup errors in the presence of motion. Nevertheless, the clinical utility of 4D-CBCT is currently limited by its extended acquisition time, degraded image quality, and slow reconstruction speed. With recent advancements, deep learning techniques have been widely used in the medical imaging field. The goal of this thesis is to develop methods that integrate deep learning into the motion compensated (MoCo) reconstruction framework to enhance the quality and efficiency of 4D-CBCT reconstruction. In the first part of the thesis, we introduce a deep learning-based artifact reduction method for preliminary reconstructed 4D-CBCT images, aiming to improve the accuracy of motion modeling and the quality of subsequent MoCo reconstruction. In the second part, we investigate the feasibility of utilizing deep learning-based groupwise registration to estimate motion models for MoCo reconstruction. Two registration models employing different training strategies are implemented and evaluated. The proposed methods demonstrate the ability to achieve fast motion modeling for MoCo reconstruction without compromising image quality, which facilitates the clinical deployment of 4D-CBCT for radiation therapy. In the final part, we validate the proposed methods on the modern CBCT imaging system (HyperSight, Varian Medical Systems) to demonstrate their performance and generalizability. Additionally, we explore the application of 4D-CBCT in cardiac imaging.
Language
English (en)
Chair
Geoffrey Hugo