<p>Magnetic resonance imaging (MRI) is one of the most versatile non-invasive diagnostic modalities for the evaluation of diseases. Despite its clinical significance, MRI acquisitions often present challenges in spatial resolution and image quality due to inherent physical constraints of the scanning systems. Improving the resolution and quality of MR images has therefore become a critical task. In this paper, we propose a method that decomposes brain MR images into 2D slice sequences, employing a second-order Markov propagation strategy to exploit the correlation between adjacent slices for deep feature refinement. By integrating the correlated features of the previous slices, we enable effective feature fusion for subsequent slices. Additionally, grid connections are introduced to enhance information flow across slices, thereby improving the model’s expressiveness. Our recurrent framework efficiently assimilates information from previous slices without increasing computational overhead. Experimental results on the IXI dataset demonstrate that our method significantly enhances the quality of medical images and achieves state-of-the-art (SOTA) performance.</p>