MD-Recon-Net: A Parallel Dual-Domain Convolutional Neural Network for Compressed Sensing MRI
Compressed sensing magnetic resonance imaging (CS-MRI) is a theoretical framework that can accurately reconstruct images from undersampled k-space data with a much lower sampling rate than the one set by the classical Nyquist-Shannon sampling theorem. Therefore, CS-MRI can efficiently accelerate acquisition time and relieve the psychological burden on patients while maintaining high imaging quality. The problems with traditional CS-MRI reconstruction are solved by iterative numerical solvers, which usually suffer from expensive computational cost and the lack of accurate handcrafted priori. In this paper, inspired by deep learning's (DL's) fast inference and excellent end-to-end performance, we propose a novel cascaded convolutional neural network called MD-Recon-Net to facilitate fast and accurate MRI reconstruction. Especially, different from existing DL-based methods, which operate on single domain data or both domains in a certain order, our proposed MD-Recon-Net contains two parallel and interactive branches that simultaneously perform on k-space and spatial-domain data, exploring the latent relationship between k-space and the spatial domain. The simulated experimental results show that the proposed method not only achieves competitive visual effects to several state-of-the-art methods, but also outperforms other DL-based methods in terms of model scale and computational cost.