Date of Award
Spring 5-2020
Degree Name
Master of Science (MS)
Degree Type
Thesis
Abstract
Image reconstruction problem, especially Magnetic Resonance(MR) Imaging reconstruction, plays an important role in real life and have a great development recently. While state-of-the-art deep learning- based image reconstruction algorithms increasingly rely on machine learning, such methods usually require ground-truth and ignore physical models, which may not be satisfied in some cases and reasonable respectively. In this master research, we introduce our visions toward practical learning-based methods and two algorithms based on the visions, for solving realistic MR image reconstruction problems: 1) an deep learning approach by directly learning artifact-free 4D motion-resolved MR images from multiple noisy MR data plagued by streaking artifacts, without the need of ground truth, and 2) a new plug-and-play (PnP) based MR image reconstruction method that systematically enforces data consistency while also exploiting deep-learning imaging priors. The prior is specified through a convolutional neural network (CNN) trained to remove under-sampling artifacts from single MR data without any artifact-free ground truth.
Language
English (en)
Chair
Dr. Ulugbek Kamilov
Committee Members
Dr. Ulugbek Kamilov Dr. Hongyu An Dr. Chien-Ju Ho
Comments
Permanent URL: https://doi.org/10.7936/emc4-jj77