Development of magnetic resonance imaging based prostate treatment planning

Radiotherapy is one of the main methods used to treat prostate cancer. Radiotherapy treatment relies on accurate planning and simulation before any radiation is administered. Currently this is mainly based on CT (computed tomography) imaging, although MR (magnetic resonance) imaging provides superior soft-tissue contrast and is therefore often used to assist with accurate organ delineation. The overall treatment planning workflow and performance can be improved if the entire workflow is solely performed using MR images. In order to achieve such MR-only treatment planning, three main challenges need to be overcome: 1) the geometric accuracy of MR images needs to be assured ; 2) the MR simulator needs to be commissioned and evaluated ; 3) electron density information required for dose calculation needs to be generated from MR images. This thesis examines each of these challenges. First, a pelvic shape phantom was used to quantify the geometric distortion arising in prostate treatment. The CT image was acquired as the gold reference and the distortion of the MR image was corrected with the vendor’s built-in algorithm. Using the image registration method, the maximum geometric distortion was reduced from nearly 8 mm to within the radiotherapy tolerance level. Second, commercial radiotherapy-dedicated equipment was implemented on the Siemens Skyra 3 Tesla MR scanner. This involved a hard flat tabletop which mimicked the flat radiotherapy treatment table, and coil mounts to lift the MR coil above the patient’s body and minimise coil-induced disagreement between the MR planning and treatment geometry. A reduction in image quality was observed on the MR simulator, but no clinically significant difference was found in the accuracy of organ delineation. Furthermore, use of the MR simulator eliminated patient positioning error associated with conventional MR scanner design and thus reduced the systematic dosimetric error. The entire workflow of MR-based planning was tested using an anthropomorphic phantom and no significant difference was found between MR- and CT-based plans. Finally, substitute (also known as synthetic or pseudo) sCT images were generated from MR images using a multi-atlas local weighted voting method. Validation was conducted on 39 patients and the sCT images were in high level agreement with the CT images. In summary, MR-based radiotherapy planning for treating prostate cancer has been thoroughly tested and evaluated in this study. This may provide an important stepping stone for the future clinical implementation.