The USC Andrew and Erna Viterbi School of Engineering USC Signal and Image Processing Institute USC Ming Hsieh Department of Electrical and Computer Engineering University of Southern California

Technical Report USC-SIPI-442

“Improved Brain Dynamic Contrast Enhanced MRI Using Model-Based Reconstruction”

by Yi Guo

August 2017

Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE-MRI) is a non-invasive technique that provides information about the delivery of blood and agents within the blood to bodily organs. In oncology imaging, DCE-MRI is used to generate tracer kinetic (TK) parameter maps to evaluate tumor severity, differentiate malignant from benign, and to evaluate response to therapy. Current DCE-MRI methods have poor reproducibility, low spatial resolution, and limited spatial coverage. The overall goal of my dissertation is to improve brain DCE-MRI by using constrained reconstruction and model-based reconstruction from under-sampled raw data (sampled below the Nyquist rate). Such techniques can provide substantially higher spatio-temporal resolution, better coverage, and improved image quality with the same scan time and contrast agent dose. Using model-based reconstruction, TK parameter maps, the end-point product of DCE-MRI, can be directly estimated from under-sampled data with improved quality and reproducibility. I evaluated a specially tailored constrained reconstruction technique for DCE-MRI in brain tumor patients. The proposed technique is able to provide high spatial resolution and whole-brain coverage via under-sampling and constrained reconstruction with multiple sparsity constraints. Conventional fully-sampled DCE-MRI and the proposed whole-brain DCE-MRI were performed on the same brain tumor patients for evaluation. The proposed method is able to provide whole-brain high-resolution DCE-MRI with improved image quality compared to conventional DCE-MRI, using only 1/30th of the raw data. These advantages may allow comprehensive permeability mapping in the brain, which is especially valuable in the setting of large lesions or multiple lesions spread throughout the brain. I also developed a model-based direct reconstruction technique for DCE-MRI, where the TK parameter maps can be directly estimated from raw-data. By comparison with a state-of-the-art indirect constrained reconstruction technique, the proposed direct approach provides improved TK map fidelity, and enables much higher acceleration up to 100×. With the prospective study, this method is shown to be clinically feasible and provide high-quality whole-brain TK maps. To address the limitation of the direct reconstruction, I proposed a model consistency constrained reconstruction for DCE-MRI. The proposed reconstruction poses the tracer-kinetic (TK) model as a model consistency constraint, enabling the inclusion of different TK solvers and the joint estimation of the arterial input function (AIF) from highly under-sampled data. Good quality TK maps and patient specific AIF can be reconstructed at high under-sampling rates up to 100×.

To download the report in PDF format click here: USC-SIPI-442.pdf (6.6Mb)