UltraMR-Enforce: A Unified Ensemble Framework for Enhancement of Ultra-Low-Field MRI

Magnetic Resonance Imaging (MRI) is a crucial non-invasive technique for visualizing brain anatomy. However, in resource-limited regions, the accessibility of high-field MRI scanners (e.g., 3T) is restricted due to equipment limitations, making it challenging to obtain high-quality images for clinical diagnosis. To address this issue, this study leverages paired high-field (3T) and ultra-low-field (64mT) MRI data to design and optimize various models, ultimately proposing an enhanced version of UltraMR-Enforce. The proposed model integrates the complementary strengths of Transformer and Mamba, further enhancing structural modeling and generalization capabilities through the weighted fusion of predictions from multiple models. Extensive experiments on the challenge dataset demonstrate that UltraMR-Enforce significantly improves the clarity and contrast of 64mT MRIs while precisely preserving critical anatomical structures, thereby further enhancing the clinical applicability and diagnostic reliability of ultra-low-field MR imaging.