Robust phase unwrapping for Dixon magnetic resonance imaging multiresolution pixel clustering and local surface fitting.

BACKGROUND

Water-fat separation plays a crucial role in clinical magnetic resonance imaging (MRI) examinations. However, the Dixon technique is encumbered by the challenge of water-fat swapping due to factors such as magnetic field inhomogeneity and low image signal-to-noise ratio (SNR). This study aimed to develop a multiresolution pixel clustering and local surface fitting (MrCLOSE) phase-unwrapping method to improve the robustness of water-fat separation in two-point Dixon MRI.

METHODS

MrCLOSE first constructs a multiresolution pyramid via acquired in-phase and out-of-phase complex MRI. It then performs hierarchical phase unwrapping with pixel clustering and local surface fitting (CLOSE) within a coarse-to-fine multiresolution framework. The performance of MrCLOSE was systematically evaluated and compared with those of B0 mapping with magnitude-based correction for bipolar two-point Dixon (B0-NICEbd), rapid opensource minimum spanning tree algorithm (ROMEO), Graph Cut, and the original CLOSE algorithm with both simulation and in vivo datasets.

RESULTS

In the simulation data, MrCLOSE achieved robust phase unwrapping under different thresholds, varying SNRs, disconnected regions, and phase changes from 4π to 10π. In the threshold-dependence simulation, CLOSE showed high error ratios of 43.71% and 94.74% at thresholds of π/10 and π/5, whereas MrCLOSE reduced these values to 0.67% and 0.11%, respectively. In additional simulations involving varying SNRs, disconnected regions, and increasing phase changes, MrCLOSE consistently maintained low error ratios and outperformed ROMEO and Graph Cut. In the 355 in vivo MR images acquired from the ankle, lumbar spine, pelvis, and knee, the overall water-fat swap ratio was 3.38% for MrCLOSE, while it was 17.75% for B0-NICEbd, 96.34% for ROMEO, 67.61% for Graph Cut, and 16.34% for CLOSE.

CONCLUSIONS

By innovatively integrating multiresolution strategies into CLOSE, MrCLOSE achieves high robustness and accuracy in phase unwrapping and provides a promising solution for phase-related MRI applications, particularly for water-fat separation in clinical scenarios.