Abstract:A deep framework for the accelerated multislice magnetic resonance imaging (MRI) reconstruction is proposed here. It exploits the non‐local interslice correlation information between adjacent slices in MRI image sequences to reconstruct high‐quality undersampled MRI images. A novel texture Transformer architecture for MRI reconstruction is designed. Such a framework enables united feature learning across target slice and adjacent slices, which encourages the exploration of deep features and transferring of accurate texture features across adjacent slices by the attention mechanism. The proposed multislice texture transformer network can be constructed in a cross‐scale way. It enables to recovery textures of MRI images via different levels, and enables to reserve more details in MRI reconstruction. Magnetic resonance imaging (MRI) is a non‐interposition imaging technique that provides rich anatomical and physiological information. Yet it is limited by the long imaging time. Recently, deep neural networks have shown potential to significantly accelerate MRI. However, most of these approaches ignore the correlation between adjacent slices in MRI image sequences. In addition, the existing deep learning‐based methods for MRI are mainly based on convolutional neural networks (CNNs). They fail to capture long‐distance dependencies due to the small receptive field. Inspired by the feature similarity in adjacent slices and impressive performance of Transformer for exploiting the long‐distance dependencies, a novel multislice texture transformer network is presented for undersampled MRI reconstruction (TTMRI). Specifically, the proposed TTMRI is consisted of four modules, namely the texture extraction, correlation calculation, texture transfer and texture synthesis. It takes three adjacent slices as inputs, in which the middle one is the target image to be reconstructed, and the other two are auxiliary images. The multiscale features are extracted by the texture extraction module and their inter‐dependencies are calculated by the correlation calculation module, respectively. Then the relevant features are transferred by the texture transfer module and fused by the texture synthesis module. By considering inter‐slice correlations and leveraging the Transformer architecture, the joint feature learning across target and adjacent slices are encouraged. Moreover, TTMRI can be stacked with multiple layers to recover more texture information at different levels. Extensive experiments demonstrate that the proposed TTMRI outperforms other state‐of‐the‐art methods in both quantitative and qualitative evaluationsions.

TRANS-Net: Transformer-Enhanced Residual-Error AlterNative Suppression Network for MRI Reconstruction.

Attention-Enhanced Convolutional Iterative Unrolling Network for Under-Sampled MRI Reconstruction

T2LR-Net: an Unrolling Network Learning Transformed Tensor Low-Rank Prior for Dynamic MR Image Reconstruction.

CRNN-Refined Spatiotemporal Transformer for Dynamic MRI reconstruction

Magnetic Resonance Image Processing Transformer for General Reconstruction

T$^2$LR-Net: An Unrolling Reconstruction Network Learning Transformed Tensor Low-Rank prior for Dynamic MR Imaging

HUMUS-Net: Hybrid unrolled multi-scale network architecture for accelerated MRI reconstruction

ReconFormer: Accelerated MRI Reconstruction Using Recurrent Transformer

Deep MRI Reconstruction: Unrolled Optimization Algorithms Meet Neural Networks

Deep Learning Based MRI Reconstruction with Transformer

An improved low‐rank plus sparse unrolling network method for dynamic magnetic resonance imaging

DCT-net: Dual-domain cross-fusion transformer network for MRI reconstruction

Accelerated MRI with Un-trained Neural Networks

GLFormer: an Efficient Transformer Network for Fast Magnetic Resonance Imaging Reconstruction.

Deep plug-and-play MRI reconstruction based on multiple complementary priors

Task Transformer Network for Joint MRI Reconstruction and Super-Resolution

TTMRI: Multislice texture transformer network for undersampled MRI reconstruction

Self-supervised Deep Unrolled Reconstruction Using Regularization by Denoising

Undersampled MR Image Reconstruction Using an Enhanced Recursive Residual Network.

MDH-Net: Advancing 3D Brain MRI Registration with Multi-Stage Transformer and Dual-Stream Feature Refinement Hybrid Network

Computationally Efficient IMplicit Training Strategy for UNrolled NEtworks (IMUNNE): A preliminary analysis using accelerated real-time cardiac cine MRI