LCB-SRNet: Advancing MRI Super-Resolution Using Local Context Block and Fusion-Enhanced Bottleneck

Swin Transformer combines the advantages of CNN and Transformer in image super-resolution reconstruction tasks and shows great potential. However, its window self-attention mechanism (WSA) is limited by a small receptive field, and it is difficult to capture a wide range of image information effectively. In addition, many deep learning super-resolution methods have high computational costs. To address these problems, this paper introduces Local Context Block (LCB) for MRI image super-resolution reconstruction. LCB expands the receptive field through a sliding window self-attention mechanism (sliding-WSA) and can more effectively restore degraded MRI image pixels. On this basis, we propose an efficient MRI super-resolution reconstruction network LCB-SRNet and fuse multi-scale features through a Fusion-Enhanced Bottleneck (FEB) to further improve the reconstruction effect. Experimental results show that LCB-SRNet performs well in MRI image reconstruction performance.