Super-resolution Reconstruction of 3T-like Images from 0.35T MRI Using A Hybrid Attention Residual Network

Magnetic resonance (MR) images from low-field scanners present poorer signal-to-noise ratios (SNRs) than those high-field at the same spatial resolution. To obtain a clinically acceptable SNR, radiologists operating use much smaller acquisition matrix scanners. Thus, current state of image quality indicates need for further research to improve systems. Strategies based on super-resolution (SR) techniques can be alternatives reconstruction. However, predetermined degradation methods embedded in these techniques, such as bicubic downsampling, seem impose performance drop when actual is different pre-defined assumption. In this study, we collected unique dataset by scanning 70 participants address problem. The anatomical locations scanned slices were 0.35T and 3T data. Low-resolution (LR) (0.35T) high-resolution (HR) (3T) pairs used data training. Herein, introduce novel CNN-based network with hybrid attention mechanisms (HybridAttentionResNet, HARN) adaptively capture diverse information reconstruct MR (3T-like images). Specifically, proposed dense block combines variant blocks extract abundant features LR images. experimental results demonstrate that our residual efficiently recovers significant textures while rendering high peak ratio (PSNR) an appealing structural similarity index (SSIM). Moreover, extensive subjective-mean-opinion-score (SMOS) proves promising clinical application using HARN.