Reparameterization lightweight residual network for super-resolution of brain MR images

As the demand for high-resolution medical images increases, super-resolution (SR) technology becomes particularly important. In recent years, SR technology based on deep learning has achieved remarkable achievements, and its application in medical images is also growing. Brain magnetic resonance imaging (MRI), a critical tool for clinical diagnosis, often suffers from artifacts caused by long scanning times or motion, compromising diagnostic reliability. While deep learning-based SR methods have significantly improved, their computational complexity and resource demands hinder real-time applications in constrained environments. To address these challenges, this paper proposes a lightweight SR MRI model based on BSRN, combined with structural reparameterization, to enhance efficiency. During training, the model employs a multi-branch structure, integrating branches into a single 3 × 3 convolution in inference, significantly reducing computational complexity and storage requirements while retaining crucial feature information. Experimental results on the IXI dataset demonstrate superior performance, with notable improvements in image clarity and detail reconstruction, especially for noisy and blurred inputs. Compared to existing methods, the proposed approach balances lightweight design and performance and has good application potential, providing new ideas for future medical image processing technology development.