Anish Lahiri1, Guanhua Wang2, Saiprasad Ravishankar3, and Jeffrey A. Fessler1
1Electrical and Computer Engineering, University of Michigan, Ann Arbor, MI, United States, 2Biomedical Engineering, University of Michigan, Ann Arbor, MI, United States, 3Computational Mathematics, Science and Engineering, and Biomedical Engineering, Michigan State University, East Lansing, MI, United States
1Electrical and Computer Engineering, University of Michigan, Ann Arbor, MI, United States, 2Biomedical Engineering, University of Michigan, Ann Arbor, MI, United States, 3Computational Mathematics, Science and Engineering, and Biomedical Engineering, Michigan State University, East Lansing, MI, United States
Our findings indicate that due to synergy between learned features, there is significant benefit to combining shallow, sparse prior-based blind learning reconstruction with deep-supervised reconstruction in MRI
Figure 5: Comparison of reconstructions for a knee image using the proposed BLIPS method versus strict supervised learning, blind dictionary learning, and zero-filled reconstruction for the 5-fold undersampling mask depicted in Fig. 2. Metrics listed below each reconstruction correspond to PSNR(in dB)/SSIM/HFEN respectively.
Figure 1: Proposed pipeline for combining blind dictionary learning and supervised learning-based MR image reconstruction.