MRI Reconstruction with Enhanced Self-Similarity Using Graph Convolutional Network

Abstract

Abstract Background Recent convolutional neural network (CNN) performs low-error reconstruction in fast magnetic resonance imaging (MRI). Most of them convolve the image with kernels and have successfully explored the local information. However, the non-local image information, which is embed among image patches that are relatively far from each other, may be lost since the convolution kernel size is usually small. We aim to incorporate a graph to represent non-local information, and improve the reconstructed images by Enhanced Self-Similarity Using Graph Convolutional Network (GCESS). Methods First, image is reconstructed into graph to extract the non-local self-similarity in the image. Second, GCESS uses graph convolution and spatial convolution to process the information in the image, so that local and non-local information can be effectively utilized. The network strengthens the non-local similarity between similar image patches while reconstructing images, making the reconstruction details more reliable. Results Experimental results on in vivo knee and brain data demonstrate that the proposed method achieves better artifacts suppression and details preservation than state-of-the-art methods, both visually and quantitatively. Under 1D Cartesian sampling with 4x acceleration (AF=4), the PSNR of knee data reached 34.19 dB, 1.05 dB higher than that of the compared methods; the SSIM achieved 0.8994, 2% higher than the compared methods. Similar results were obtained for the reconstructed images under other sampling templates as demonstrated in our experiment. Conclusions The proposed method successfully construct a hybrid graph convolution and spatial convolution network to reconstruct images. Along with the network training, the non-local self-similarities are enhanced, and will benefit the image details reconstruction. Experiments demonstrate that the proposed method outperforms the state-of-the-art reconstruction method in suppressing artifacts, as well as in preserving image details.