Study on MRI Image Denoising Method Based on Self-Supervised Deep Convolutional Network

Image denoising technique is of great significance in improving image quality. In this paper, we propose an MRI image denoising method based on self-supervised deep convolutional network, which adopts the encoder-decoder structure to design the network architecture, and innovatively combines the multi-scale feature fusion, null convolution, and attention mechanism for noise suppression. With the self-supervised learning strategy of 'noise2noise', the method can be trained without the need of paired clean images. The method is experimentally verified to outperform existing algorithms in terms of denoising effect and structure preservation.