The Medical Image Denoising Method Based on the CycleGAN and the Complex Shearlet Transform

Abstract

Medical image denoising plays an important role for the noise in the medical images can reduce the visibility, thereby affecting the diagnostic results of the doctors. Although good results have been achieved by the well-known deep learning-based denoising methods for their strong ability of learning, the loss of structural feature information and the well preservation of the edge information have not attracted considerable attention. To deal with these problems, a novel medical image denoising method based on the improved CycleGAN and the complex shearlet transform(CST) is proposed. The CST is used to construct the generator to embed more feature information in the training process and the denoising process is modeled to adversarial learn the mapping between the noise-free image domain and the noisy image domain. With the mechanism of the recurrent learning from the CycleGAN, the proposed method does not need the paired training data, which obviously speeds up the training and is more convenient than other classical methods. By comparing with five state-of-the-art denoising methods, experiments on the open dataset fully prove the accuracy and efficiency of the proposed method in terms of the visual quality and the quantitative PSNR, SSIM, and EPI.

A Novel and Efficient Point Cloud Registration by using Coarse-to-Fine Strategy Integrating PointNet

“…Firstly, a test on the learned model by using the open ModelNet40 dataset [36][37][38][39][40] is implemented. The intermediate results in the iterations are shown in Fig.…”