Optimal integration of MRI in Radiation Oncology is hindered by the lack of methods that harvest the significant prior knowledge available to sample the anatomy, biological status, and physiologic motions of individual patients. While some generic image acquisition methods take advantage of non-specific low rank structure of human MR signals to achieve some modest acceleration, the wealth of specific prior knowledge, from both the population of similar patients as well as the specific patient, has yet to be effectively tapped to guide optimal treatment planning, positioning, and monitoring. We hypothesize that biological, morphological, and motion models of the patient can be accurately derived from a limited number of samples aided by prior knowledge. These advances will allow us to reduce scan times dramatically (to less than 10% of conventional scanning) for morphological imaging, support efficient biological imaging for high order diffusion modeling and create hierarchical motion-frozen image volumes of abdominal patients that simultaneously provide breathing, GI contraction, and potentially cardiac motion models with probability density functions that can be used to estimate the impact of intrafraction motion on treatments and eventually select local navigators for real-time monitoring of specific regions that are most sensitive to motion-related impacts on delivered doses to targets or organs at risk. We will investigate this hypothesis by developing a prior knowledge-based compressed sensing method to reconstruct densely sampled DW attenuation curves from sparsely sampled ones; performing principal component analysis of previously scanned FLAIR, contrast-enhanced T1-weighted and Diffusion-Weighted image volumes to support sparse sampling in k-space for anatomic imaging and in b-values for diffusion imaging; investigating potential gains in acceleration of imaging by combining a patient-specific prior with population-derived principal components of structure and diffusion; modeling breathing and peristaltic motion. Finally, we will develop and implement scanning sequences based on the modeled methods for subsampling b-values and anatomy. By these methods, we expect to provide efficient anatomic and high order diffusion imaging, as well as introduce means to automatically extract hierarchical motion models of the patient for use in treatment planning and future support of treatment monitoring. Relevance to PAR 18-484 (for the NCI): This investigation seeks to improve both the efficiency as well as the efficacy of precision radiation therapy for patients with GBMs, other intracranial targets as well as intrahepatic tumors. As Radiation therapy is part of the standard armamentarium of care options for these patients, this research falls within the purview of the NCI.
Magnetic Resonance imaging is rapidly advancing as a primary tool for guiding Radiation Therapy. With routine availability of MR Simulators as well as significant expansion of MR-guided treatment systems, the Radiation Oncology community will benefit greatly from methods to improve the speed and discriminatory power of Magnetic Resonance to monitor the shape, motion, and biological structure of tumors and normal tissues efficiently and accurately. Taking advantage of two concepts, both the lack of significant complexity in descriptions of shape, biology and motion, as well as prior knowledge from both the individual patient as well as populations of similar patients, we will produce highly efficient maps of diffusion for tumor identification and monitoring, patient shape for very rapid scanning both for planning as well as guiding treatments precisely, and finally images that can freeze the patient at selectable combinations of breathing and gastrointestinal motion states, as well as model these motions for more accurate estimation of delivered radiation doses and selection of treatment monitoring methods.
Showing the most recent 10 out of 14 publications
