Boosting Magnetic Resonance Image Denoising With Generative Adversarial Networks

Denoising plays an important role in the Magnetic Resonance Imaging (MRI) applications for medical diagnosis. MRI images usually contain undesired noises which would negatively affect exactitude of pathological Recently, many models denoising have been developed from deep learning networks. In this paper, we propose a novel image method using conditional Generative Adversarial Networks (GANs). Specifically, Convolutional Neural Network (CNN) is utilized as discriminator process to distinguish whether pair obtained GAN real consists noisy and noise-free or fake which, on other hand, contains denoised image. our design, convolutional encoder-decoder networks-based generator used remove noise much possible. The whole architecture trained by adversarial learning. Experiments both synthetic clinical datasets are conducted. When tested T1w with 10% level, performed better terms reaching high structural similarity index (SSIM) at 0.9489 while that next best was only 0.7485. Moreover, when level increased 1% 10%, more stable SSIM dropped about 3.2% 23.7%. Simulation results demonstrate proposed robust outperforms conventional methods preservation anatomical structures defined contrast.