A Dilated Convolution-based Denoising Network for Magnetic Resonance Images

Magnetic Resonance Imaging (MRI) are typically corrupted with random noise. This type of noise exhibits the characteristics of Rician distribution in MRI scans. Noise in MRI scans degrades the accuracy of manual and computerized inspection of diseases. Therefore, denoising of MRI images is an indispensable process before the clinical examination of any disease. In this article, we present a novel denoising neural network for MRI images. The proposed network contains a set of dilated convolutions for Rician noise removal. We have used hybrid dilated convolutions to overcome the gridding problem in the network. The residual learning scheme has also been utilized using a set of skip connections. A substantial amount of supervised MRI data has been developed for end-to-end training of the proposed network. Extensive experiments have been performed on synthetic and real MRI datasets to study the effectiveness of the proposed method. The experimental observations indicate that our method not only achieves promising performance but also retains prominent image information effectively.