A Speckle Denoiser: Deep Convolutional Neural Network for MR Image Denoising

Image de-noising is used to enhance the quality of images because while acquiring images they are degraded by different noises. Medical images get corrupted by gaussian and speckle noises in general. In this paper, Noisy image is obtained by adding either Gaussian noise or speckle noise of range 5 to 50. For MR brain tumor image de-noising purpose, deep learning models like pretrained DnCNN, and proposed deepCNN are considered. The performance of deep learning models are evaluated in terms of parameters like PSNR, SSIM, MSE, and MAE. The proposed DeepCNN model gives better performance in achieving denoising than other method due to usage of batch normalization, ReLU, and 17 convolutional layers so that the training process gets speeded up, the denoising performance is boosted and over fitting problem is also avoided by doing augmentation of the datasets.