Randomized trial demonstrates accurate MRI super-resolution via deep learning, indicating strong clinical potential.
Key Points
This study aims to develop a deep learning framework for MRI super-resolution that enhances image quality while minimizing computational demands.
Proposed a hybrid model integrating multi-head selective state-space models and lightweight channel multilayer perceptron.
Utilized 2D patch extraction with hybrid scanning strategies to capture long-range dependencies.
Trained and evaluated on two datasets: 7T brain T1 MP2RAGE maps (142 subjects) and 1.5T prostate T2w MRI (334 subjects).
Achieved highest structural similarity (SSIM) and peak signal-to-noise ratio (PSNR) for the 7T brain dataset, with statistically significant improvements over all baselines.
For the prostate dataset, similarly outperformed competing methods in SSIM and PSNR metrics.
Model maintained exceptional efficiency with only 0.9 million parameters and 57 GFLOPs, representing significant reductions compared to existing methods.