UCSF

Incidental discoveries of renal cell carcinomas, the most common type of kidney cancer, have increased in recent years due to improved imaging technologies. It is crucial to be able to characterize the pathology and tumor grade of discovered renal masses to optimize treatment planning. This problem will be approached in this dissertation using advanced metabolic MRI methods and CT data collection.Hyperpolarized [1-13C]pyruvate (HP C13) MRI has emerged as a method of imaging metabolic pathways in cancer, including kidney cancer. Yet further acquisition and quantification improvements are needed to optimize clinical relevance. To improve HP C13 MRI acquisition in preclinical systems, a spectral-spatial echo planar imaging (EPI) sequence is proposed and compared to using a chemical shift imaging (CSI) acquisition. The EPI sequence is found to reduce partial volume effects and expedite acquisition. To improve upon quantification of HP C13 MRI, a novel pharmacokinetic model for balanced steady state free precession (bSSFP) acquisitions is proposed to fit apparent rate constants, kPL and kPB. The fit rate constant maps are compared to results from previous pharmacokinetic models in paired preclinical and clinical datasets. Additionally, a U-Net is trained to estimate kPL maps from HP C13 MRI data to take advantage of spatial relationships in the data. A novel anatomic HP C13 MRI brain phantom is introduced as training data for the U-Net. The U-Net is further finetuned with in vivo datasets. The U-Net predicted rate constant maps are compared to using a pharmacokinetic model to fit maps. Finally, a 500+ 3D multiphase renal tumor CT dataset is described to increase available data examples of renal tumors for better performance of data-driven approaches in renal tumor characterization.