UCLA

Magnetic resonance imaging-only (MRI-only) simulation for external beam radiation therapy treatment planning of prostate cancer has seen increased clinical use. The use of a single imaging modality for simulation imaging brings benefits to radiation therapy workflows such as the elimination of systematic positional errors associated with multimodal image registration during treatment planning. However, several challenges remain for the widespread clinical adoption of MRI-only simulation imaging for radiation therapy such as the lack of robust pre-treatment alignment methods and dedicated quality assurance testing equipment.

In the MRI-only simulation imaging workflow, synthetic computed tomography (CT) images are created for a variety of uses including providing tissue electron density information for dose calculations. Synthetic CT image generation algorithms are typically trained using patient data and are highly sensitive to human tissue contrast and geometry. Most institutions that treat patients with MRI-only simulation images cannot use commercially available phantoms to quality assurance test processes such as synthetic CT image generation. This is because most commercially available phantoms do not mimic human tissue geometry and tissue imaging characteristics for both MRI/CT modalities. The absence of MRI/CT compatible end-to-end quality assurance testing instruments could potentially lead to systematic errors in treatments using MRI-only simulation imaging because of the lack of imaging and dosimetric benchmarking standards.

Studies on the commissioning of MRI-only simulation imaging for radiation therapy of prostate cancers have recommended the use of intraprostatic fiducial markers for pre-treatment patient positioning and alignment. However, fiducial markers appear as dark signal voids in MRI and are challenging to manually localize without the aid of CT imaging. Other intraprostatic objects such as calcifications produce similar signal voids to fiducial markers in MRI images. There is currently no consensus on the optimal fiducial marker or MRI sequence to detect fiducial markers with a high level of sensitivity and specificity in MRI-only simulation images. Additionally, there are no clinically available automatic marker detection workflows available to aid in the clinical transition to MRI-only simulation imaging.

This thesis presents work undertaken to meet the challenges of the clinical development of MRI-only simulation imaging for radiation therapy of prostate cancers. In the presented work, the author describes the development of a novel system of multimodal tissue mimicking materials for MRI and CT imaging. The aforementioned system of materials was adapted into a novel 3D-printed anthropomorphic phantom for quality assurance testing of MRI-only simulation procedures. To address the issues with patient positioning and alignment, a human and phantom study was conducted to quantitatively evaluate the optimal fiducial marker and MRI sequence for patients receiving MRI-only radiation therapy simulation imaging. Finally, an automatic deep-learning based fiducial marker detection algorithm is presented to aid with the clinical transition of CT-based to MRI-only radiation therapy simulation workflow.