MS2CAM: Multi-scale self-cross-attention mechanism-based MRI super-resolution

Abstract

In magnetic resonance imaging (MRI), it is challenging to achieve both real-time imaging and high image quality due to its unique imaging modality. Low-resolution (LR) often accompanies real-time MRI, making super-resolution (SR) techniques essential for enhancing image quality in real-time MRI. This paper proposes a Multi-Scale Self-Cross-Attention Mechanism (MS²CAM) for MRI SR tasks, where concrete and abstract features are effectively fused to improve SR performance. Our model demonstrates consistent performance improvements of 1–2% over state-of-the-art methods across various degradation scenarios. Visual results also reveal finer detail restoration, verifying MS²CAM’s effectiveness. Extensive experimental results confirm that MS²CAM achieves superior quantitative and visual performance in MRI SR tasks, establishing it as a leading solution in this domain.