Multi-scale Inter-frame Information Fusion Based Network for Cardiac MRI Reconstruction

Accelerated cine MRI reconstruction from under-sampled data is paramount in clinical diagnosis. Nonetheless, the existing method falls short in fully harnessing inter-frame information, thereby impeding its overall reconstruction performance. This paper presents a novel multi-scale inter-frame information fusion strategy, aimed at extracting and leveraging the multi-scale features from adjacent multi-frame data to guide the reconstruction process more comprehensively. The proposed framework incorporates several specific encoders for feature extraction from each frame, followed by an information fusion block that effectively combines the multi-scale features of multiple frames. This ensures the effective utilization of supplementary information from multiple frames at varying scales. Moreover, the fused inter-frame information is also utilized in subsequent refinement blocks to perform feature enhancement and guide the reconstruction. Consequently, the introduced multi-scale inter-frame information fusion strategy not only enhances the overall reconstruction performance but also demonstrates high efficiency. Experimental results show that the method can obtain competitive reconstruction results and the metrics of SSIM, NMSE and PSNR can reach 0.9299, 0.0098 and 35.43 respectively averaging over all validation datasets.