Deep Learning Approaches for Banding Artifact Suppression in Single Phase-Cycled bSSFP

Motivation: bSSFP is widely used in clinical applications for its high speed and high signal-to-noise ratio (SNR). However, off-resonance effects cause banding artifacts, requiring multiple phase-cycled datasets, adding complexity and time. Goal(s): This study aims to suppress bSSFP banding artifacts using deep learning, achieving high-quality, artifact-suppressed images with only single phase-cycling dataset. Approach: By leveraging a single phase-cycling angle, we developed U-Net and Restormer models, known for their effective image related tasks, to suppress banding artifacts. Results: Deep learning models effectively suppressed banding artifacts, achieving better results with less data than conventional two-angle approaches. Restormer showed better performance than U-Net. Impact: With only one phase-cycling data, deep learning models, particularly Restormer, effectively suppress bSSFP banding artifacts, resulting in high-quality, artifact-suppressed images that maintain the speed and SNR benefits of bSSFP, facilitating effective clinical judgments and future research.