Development of Anatomical Patient Models to Facilitate MR-only Treatment Planning

点击翻译按钮获取中文摘要

Accurate delineation of targets and organs at risk for radiation therapy planning (RTP) remains a challenge due to the lack of soft tissue contrast in computed tomography (CT), the standard of care imaging for RTP. Radiation Oncology has addressed this limitation by registering magnetic resonance images (MRI) to CT datasets to take advantage of the superior soft tissue contrast afforded by MRI. MRI brings considerable value to RTP by improving delineation accuracy which, in turn, has enabled dose escalation to improve local control while maintaining or reducing normal tissue toxicities. However, the current integration of MRI as an adjunct to CT has significant drawbacks as it requires image registration and contour transfer between datasets. This process introduces systematic geometric uncertainties that persist throughout treatment and may compromise tumor control. Thus, we propose to translate MR-only RTP into clinical use, with the ultimate goal of improving patient outcomes accomplished via improved treatment plan design. MR-only RTP will eliminate redundant CT scans (reducing dose, patient time, and costs), streamline clinical efficiency, entirely circumvent registration uncertainties, and fully exploit the benefits of MRI for high-precision RTP. Yet, MRI is not routinely used alone for RTP, largely due to its known spatial distortions, lack of electron density, and inability to segment the bone needed for online image guidance and electron density mapping for dose calculation. The central hypothesis is that the innovative technologies that our multi-disciplinary academic/industrial (Henry Ford Health System/Philips Healthcare) collaboration develop will yield geometrically accurate patient models built from MRI data across several platforms/field strengths with CT-equivalent densities that can be used in confidence throughout the entire RTP workflow. In Aim 1, we will perform geometric distortion corrections, determine distortion variability with changing anatomy, benchmark the results in a novel modular phantom, and develop an image processing toolkit. In Aim 2, we will fully automate MR image segmentation in the brain and male/female pelvis to yield accurate synthetic CT patient models derived from novel MRI sequences, including provisions for metal implants, and benchmark the results in phantom. In Aim 3, we will conduct end-to-end testing to characterize the uncertainties in the MR-only RTP workflow. We will perform a virtual clinical trial of MR-only RTP for brain and male/female pelvis and compare to the standard of care. Final translation will include developing physician-physicist practice guidelines, end-user validation of all translational steps, and dissemination of image processing tools into the Radiation Oncology community. This research will systematically address the major challenges limiting MR-only RTP and lay the groundwork for multi-institutional clinical trials across MRI platforms. It will support future work related to MR-guided RT, functional MRI for biologically adaptive RT, and focal RT to areas of high tumor burden.