Author

Zeheng Li

Graduation Semester and Year

2021

Language

English

Document Type

Thesis

Degree Name

Master of Science in Computer Science

Department

Computer Science and Engineering

First Advisor

Junzhou Huang

Abstract

Low-dose computed tomography (LDCT) has raised highly attention since the counterpart, full-dose computed tomography (FDCT), brings potential ionizing radiation influence to patients. However, LDCT still suffers from several issues such as relatively higher noise level, which limits its uses in practical applications. To improve LDCT image quality, conventional denoising methods, such as KSVD and BM3D, are first introduced to suppress noise in low-dose images. These methods, however, works under assumptions that are not robust to various data. In this paper, we conduct an extensive research on deep learning based denoising method in LDCT images. We mainly base on Generative-Adversarial Network (GAN) variants, such as CycleGAN, IdentityGAN and GAN-CIRCLE, and compare their performance in low-dose image denoising. Compared to supervised deep learning methods, these GAN based methods effectively learn image translation from the low-dose domain to the full-dose (FD) domain without the need of aligning FDCT and LDCT images, which is usually required in other methods for domain translation. Experiments on real and synthetic patient CT data show that these methods can achieve comparable peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) to, if not better than, the other state-of-the-art denoising methods. Among CycleGAN, IdentityGAN, and GAN-CIRCLE, the later achieves the best denoising performance with the shortest wall clock time. In addition, we use GAN-CIRCLE to demonstrate that the increasing number of training patches and training patients can improve denoising performance. Finally, two non-overlapping experiments, i.e., no counterparts of FDCT and LDCT images in the training data, further demonstrate the effectiveness of unpaired learning methods. This work paves the way for applying unpaired GAN based methods to enhance LDCT images without requiring aligned FD and low-dose images from the same patient.

Keywords

Deep learning, GAN, CT denoising

Disciplines

Computer Sciences | Physical Sciences and Mathematics

Comments

Degree granted by The University of Texas at Arlington

29801-2.zip (3600 kB)

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.