Super-resolution reconstruction of Knee MRI at 3T and 7T based on the Hybrid Attention Transformer

Motivation: Enhancing lower-resolution images from 3T/7T MR to match high-resolution images of 7T using HAT algorithm could combine the diagnostic benefits of 7T for knee disorders with the availability of 3T. Goal(s): To investigate the performance of HAT algorithm in super-resolution reconstruction for improving lower-resolution images from 3T and 7T MR. Approach: For the 7T knee image task, we used MSE-based HAT with 0.8mm and 0.4mm datasets, then fine-tuned a GAN-based 3T restoration model on 3T-7T pairs using the pre-trained 7T SR model. Results: The MRI-based knee HAT algorithm enhanced the image quality and spatial resolution of lower-resolution imaging from 3T and 7T MR. Impact: This study demonstrates the potential of accurately identifying and characterizing peripheral nerve pathology in the extremities utilizing 3D DESS and PD-TSE FS sequences at 7T MR.