Description
Lung cancer is the second most diagnosed and the leading cause of cancer death in the US. It accounts for nearly 15% of all new cancer cases, and about 28% of all cancer deaths. Unfortunately, lung cancer radiotherapy is associated with a poor clinical outcome. Thus, the need for an aggressive radiation therapy regimen, that is, involving fewer fractions and higher radiation doses per fraction to tumor targets while increasing healthy tissue sparing, is evident for increasing local control rates and clinical outcome. Stereotactic body radiation therapy (SBRT) is among current state of the art techniques that fill this need by providing highly conformal, high-dose radiation doses to cancerous tumors. Such techniques as SBRT rely on state of the art imaging systems to provide precise localization of tumor targets, as well as critical organs at risk, throughout all stages of the radiotherapy process from treatment simulation and planning, and throughout the radiation delivery. The main source of uncertainty in radiation delivery of lung cancer is due to the respiration-induced deformation of the thoracic anatomy during imaging/treatment. Therefore, the four-dimensional computed tomography (4DCT) imaging is a crucial step in the design of a highly conformal SBRT plan. 4DCT captures the anatomy at multiple stages of the respiratory cycle. However, the current SBRT plans are based on a single aggregate CT set, such as the maximum intensity projection (MIP) or the average intensity projection (AIP) CT images, which is derived from a 4DCT dataset and represents a motion encompassing CT image on which treatment planning is based. However, this imaging method, while saves time, presents a limitation on SBRT since neither MIP nor AIP CT images correctly represent the moving anatomy. The resulting planned dose and actual delivered dose may or may not be substantially different depending on each patient case. Deformable image registration (DIR) is an image processing technique that calculates the relative motion magnitude and direction of each image voxel between a corresponding two images of the same anatomy. The result can, in principle, be used to correctly account for the motion-induced errors in dose calculations, and thus provides means to verify the accuracy of radiation dose delivery, known as 4D planning. The goal of this thesis is to pursue the viability of this verification process. The two well-known DIR algorithms were studied and implemented: (1) Horn-Schunck's optical flow, and (2) Demons algorithms. In this thesis, a representative two lung SBRT plans were re-calculated based on the DIR between all 4DCT image phases, and the resulting "4D doses" were compared to the original planned doses. Results showed that the current MIP-based SBRT planning doses did not significantly differ from the full 4D plan doses. Moreover, it was shown that the optical flow algorithm is faster and more accurate than the Demons algorithm. Further studies are needed to validate our groundbreaking work in the future.
