关键词: 图像重建, 欠采样图像, 复卷积, 双域学习, 深度学习

Abstract: At present, most accelerated Magnetic Resonance Imaging (MRI) reconstruction algorithms reconstruct undersampled amplitude images and use real-value convolution for feature extraction, without considering that the MRI data itself is complex, which limits the feature extraction ability of MRI complex data. In order to improve the ability of feature extraction of single slice MRI complex data, and thus reconstruct a single slice MRI image with clearer details, a Complex Convolution Dual Domain Cascade Network(ComConDuDoCNet) was proposed. The original undersampled MRI data was input into the proposed method, and Residual Feature Aggregation (RFA) blocks were used to alternately extract the dual domain features of the MRI data, ultimately reconstructing the Magnetic Resonance(MR) image with clear texture details. Complex convolution was used as a feature extractor for each RFA block. Different domains were cascaded through Fourier transform or inverse transform, and data consistency layer was added to achieve data fidelity. A large number of experiments were conducted on publicly available knee joint datasets, and the comparison results with the Dual-task Dual-domain Network (DDNet) under three different sampling masks with a sampling rate of 20% show that under the two-dimensional Gaussian sampling mask, the proposed algorithm's Normalized Root Mean Square Error (NRMSE) index decreases by 13.6%, the Peak Signal-to-Noise Ratio (PSNR) index increases by 4.3%, and the Structural SIMilarity index (SSIM) index increases by 0.8%; Under the Poisson sampling mask, the NRMSE index of the proposed algorithm decreases by 11.0%, the PSNR index increases by 3.5%, and the SSIM index increases by 0.1%; Under the radial sampling mask, the NRMSE index of the proposed algorithm decreases by 12.3%, the PSNR index increases by 3.8%, and the SSIM index increases by 0.2%. The experimental results show that ComConDuDoCNet, combined with complex convolution and dual domain learning, can reconstruct MR images with clearer details and more realistic visual effects.

Key words: image reconstruction, undersampled image, complex convolution, dual domain learning, deep learning