DS-COFFEE-Net: Dual-Stage Correspondence-Based Feature Fusion Edge Enhanced Network for Super-Resolution of CEST-MR Images

This paper proposes a deep-learning-based approach for super-resolution (SR) in chemical exchange saturation transfer magnetic resonance imaging (CEST-MRI): Dual-stage COrrespondence-based Feature Fusion Edge Enhanced Network (DS-COFFEE-Net). Two sequential stages comprise the DS-COFFEE-Net: stage-A is a conventional and widely verified RDN network, and stage-B is a novel COFFEE-net with a COFF-subnet and an EE-subnet. The COFF-subnet with the transposed transformer is exploited to capture the inner-correspondence among images in the frequency dimension (F-dim), whereas the EE-subnet improves the output image details. Our model outperforms other start-of-the-art (SOTA) methods in tests on the CEST-MRI dataset, showing good potential in clinical applications.