Low dose CT image denoising using a coupled generative adversarial network with unpaired data
Principal Investigator
Name
Fei Chen
Degrees
Ph.D.
Institution
Xidian University
Position Title
Dr.
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-1017
Initial CDAS Request Approval
Feb 21, 2023
Title
Low dose CT image denoising using a coupled generative adversarial network with unpaired data
Summary
Lowering the radiation dose of computed tomography (CT) scan inevitably degenerates the quality of the reconstructed image which may hinder the medical diagnosis. Recently, many deep learning based low dose CT (LDCT) denoising methods have been proposed and provide superior performance as compared to the model-based iterative methods. However, existing deep learning based LDCT denoising approaches require the training sample to be paired LDCT and normal dose CT (NDCT) images. In reality, such paired samples are particularly difficult to obtain. On the other hand, a large number of unpaired images are easily obtained. In this work, a coupled generative adversarial network (GAN) structure will be used to learn a LDCT denoising model by utilizing the unpaired LDCT and NDCT images. In addition, multimodal learning is performed in combination with pathological images as well as CT images for the segmentation task of lung cancer.
Aims
We would like to have more LDCT data, together with our existing paired data, to train a better LDCT denoising model to see if the denoising algorithm can help improve the model's ability to predict lung cancer. In addition, multimodal learning is performed using LDCT and pathology images to improve segmentation performance.
Collaborators
Fei Chen, PhD, Xidian University