End-to-End Medical Image Denoising via Cycle-consistent Generative Adversarial Network
Chenggeng Yan, Hu Chen, Yang Zhao · 2021
Since the radiation of X-ray is detrimental to patients, low-dose computed tomography (CT) has been developed in medical field. Nevertheless, it may degrade the quality of CT images and affect the clinical diagnosis. To address this issue, various advanced low-dose CT denoising methods have been developed to improve the quality of CT images. The majority of these methods require well-matched low-dose CT and normal-dose CT images for the training. These methods have made much progress in low-dose CT denoising field. However, it is not practicable in medical field to acquire well-paired images in medical field which restricts the application of these methods. To deal with this matter, we propose an improved end-to-end CT image denoising without aligned low-dose and normal-dose CT images inspired by Cycle-consistent Generative Adversarial Network. This method can learn the data distributions of low-dose CT and normal-dose CT in order to build a mapping from low-dose CT images to normal-dose CT images without paired CT images. Additionally, the perceptual loss is integrated into the proposed method to enhance the visual quality of denoised images. The experiments on the real clinic CT dataset show our methods can not only suppress noise but also preserve structural information.CCS-Applied computing~Computers in other domains~Computing in government~E-government