Deep Learning based Phase Correction with Noise and Artifacts Removal for MERGE

Motivation: Multiple Echo Recombined Gradient Echo (MERGE) images are inherently complex-valued, and motion, field inhomogeneities, etc. could cause echo-to-echo background phase variations. Filter-based phase correction often results in signal cancellation. Goal(s): To remove echo-to-echo phase variations for complex echo combination and improve the in-plane resolution and SNR of complex combined imageApproach: We used a deep-learning-based phase correction to improve complex echo combination and apply AIR Recon DL to further improve the in-plane resolution and SNRResults: Deep learning based phase correction minimized signal cancellation and enabled robust complex echo combination With AIR Recon DL, MERGE images showed improved resolution and SNR. Impact: With improved image quality, it could improve the visualization, segmentation and measurement of tissue of interest, improving diagnosis, treatment response monitoring, etc.