Phase Inconsistency Problem of VarNet Reconstruction with External Calibration Data

Motivation: E2E VarNet is a powerful deep learning reconstruction method for MRI, but struggles when using external calibration data due to phase inconsistencies, particularly in applications like SMS and EPI. Goal(s): To address the phase inconsistency problem in E2E VarNet, improving reconstruction quality when using external calibration data. Approach: Various solutions were explored, including fine-tuning E2E VarNet with phase-inconsistent calibration data, applying ESPIRiT for coil sensitivity mapping, and using GRAPPA to generate phase-consistent calibration data. Results: Fine-tuning with phase-inconsistent calibration and ESPIRiT showed minimal improvement. GRAPPA-based calibration provided the best results, reducing noise and improving reconstruction accuracy. Impact: The proposed GRAPPA-based workaround solution to mitigate phase inconsistency improves the ability of E2E VarNet to utilize external calibration data, enhancing reconstruction quality and expanding its potential applications in research.