Improvements of Reconstruction Performance in a Simultaneously Multislice Imaging Using Deep Learning-based Image Separation

Motivation: A new simultaneous multi-slice imaging (SMS) that does not use the sensitivity of receiver RF coils has been proposed. Image quality could be improved with the use of an appropriate network. Goal(s): Our goal is to improve the quality of reconstructed images in proposed SMS. Approach: mage reconstruction by SARA-GAN, SwinMR and 3D U-Net were tested and the results were compared with our previous work using U-Net. Results: Simulation experiments showed that higher PSNRs were obtained by SwinMR, and the best LPIPS was achieved by SARA-GAN, showing a significant improvements of image quality in appearance. Impact: The image quality of Deep learning-based SMS has been greatly improved.