Low-Dose CT Image Denoising with Improving WGAN and Hybrid Loss Function.

Zhihua Li,Qiwei Xing,Weili Shi,Huamin Yang,Wei He,Zhengang Jiang,Yu Miao,Reza Khosrowabadi

doi:10.1155/2021/2973108

Abstract

The X-ray radiation from computed tomography (CT) brought us the potential risk. Simply decreasing the dose makes the CT images noisy and diagnostic performance compromised. Here, we develop a novel denoising low-dose CT image method. Our framework is based on an improved generative adversarial network coupling with the hybrid loss function, including the adversarial loss, perceptual loss, sharpness loss, and structural similarity loss. Among the loss function terms, perceptual loss and structural similarity loss are made use of to preserve textural details, and sharpness loss can make reconstruction images clear. The adversarial loss can sharp the boundary regions. The results of experiments show the proposed method can effectively remove noise and artifacts better than the state-of-the-art methods in the aspects of the visual effect, the quantitative measurements, and the texture details.

Highlights

During recent years, the computed X-ray tomography (CT) has been one of the important practical imaging methods, which has been widely utilized in medical diagnosis
To show the capacity of our proposed denoising SSWGAN for LDCT image, four real clinical CT image datasets were applied in our study in order to avoid overfitting problem
We describe the advantages of our algorithm framework in two ways: (1) compared with other widely used traditional LDCT denoising methods and (2) compared with the latest LDCT denoising methods based on GAN

Summary

Introduction

The computed X-ray tomography (CT) has been one of the important practical imaging methods, which has been widely utilized in medical diagnosis. With the widely use of medical CT, the potential risk of ionizing X-ray radiation to patients has aroused public concern [1, 2]. The common method to reduce noise is filtering. It is an ill-posed and challenging problem [3,4,5]. Different network architectures and loss function may have a profound impact upon the learning process of the network. According to literature [8], the complexity of the denoising model is determined by the network architecture, and the loss function is related to what the denoising model learns from images and data

Objectives

Results

Conclusion