Author

Shiwei Zhou

ORCID Identifier(s)

0000-0003-3984-8917

Graduation Semester and Year

2023

Language

English

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Physics and Applied Physics

Department

Physics

First Advisor

Mingwu Jin

Abstract

Medical imaging plays a crucial role in modern healthcare, serving as a vital component in the realms of diagnosis and treatment. It encompasses a broad spectrum of techniques and technologies aimed at visualizing the internal structure, physiology and bio-chemical processes inside the human body. Medical imaging has revolutionized medical practice by enabling doctors to diagnose diseases and monitor treatments without resorting to invasive procedures. Computed Tomography (CT) is an important tool of medical imaging. A CT scan employs computer-processed combinations of multiple X-ray images taken from different angles to generate cross-sectional images, providing significantly more detailed structural information compared to 2D X-rays. However, CT relies on ionizing radiation and its image quality can deteriorate due to patient motion and reduced imaging dose. In this work, we aim to improve CT image quality (at lower radiation dose) using advanced methods ranging from traditional modeling to deep learning. In this work, we first developed a general simultaneous motion estimation and image reconstruction (G-SMEIR) method for 4D cone-beam CT (CBCT) to capture and model lung motion for radiation therapy (Chapter 2). It can overcome the local trapping problem of motion estimation and achieve better 4D CBCT image quality and motion tracking for lung tumors. Secondly, we developed several deep learning methods for CT denoising: cycle generative and adversarial network (CycleGAN) and RecycleGAN for unpaired single low-dose CT image denoising and unpaired low-dose CT image sequences denoising, respectively (Chapter 3), and texture transformer for super-resolution (TTSR) for low-dose CT (Chapter 4). These methods yield unprecedented denoising performance compared to other state-of-the-art denoising methods. This dissertation work not only provides multiple tools to address important issues in CT, but also demonstrates that advanced modeling and deep learning methods are effective in solving challenging problems in medical imaging.

Keywords

CycleGAN, RecycleGAN, Denoising, Multi-phase CT angiography (MP-CTA), Medical imaging, CT super-resolution, Texture transformer super-resolution (TTSR), Generative adversarial network (GAN), GAN with cycle-consistency (GAN-CIRCLE), Low-dose CT, 4D reconstruction, Multi-frame reconstruction with parametric motion model (MF-PMM), General simultaneous motion estimation and image reconstruction (G-SMEIR), Motion estimation, Cone-beam computed tomography (CBCT)

Disciplines

Physical Sciences and Mathematics | Physics

Comments

Degree granted by The University of Texas at Arlington

31755-2.zip (11254 kB)

Included in

Physics Commons

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.