A robust MRI water-fat separation algorithm based on field map and fat fraction map smoothness.

Area of Science:

• Medical Imaging
• Biophysics
• Image Processing

Background:

• Magnetic resonance (MR) water-fat separation is crucial for tissue characterization and fat quantification.
• Robust separation is challenging due to background noise and magnetic field inhomogeneities.

Purpose of the Study:

• To enhance robustness in challenging MR imaging, including low field strengths and large fields of view (FOV).
• To leverage the inherent spatial smoothness of field maps and fat fraction maps for improved accuracy.

Main Methods:

• Proposed the Field Map and Fat Fraction Map Smoothing (FMFMS) algorithm, incorporating multi-peak fat modeling, T2* decay correction, and field inhomogeneity compensation.
• Utilized a two-stage optimization framework for nonlinear parameter estimation and least-squares for water/fat amplitudes.
• Integrated spatial smoothness of field and fat fraction maps, identifying erroneous voxels via local field discontinuities and Local Polynomial Surface Fitting (LPSF).

Main Results:

• FMFMS outperformed existing methods (GOOSE, FFM) on ISMRM 2012 Challenge, phantom, and in vivo datasets.
• Achieved 99.5% average accuracy on ISMRM Challenge cases (excluding #3) with minimal artifacts.
• Demonstrated strong performance in low-field (0.35T) and large-FOV (1.5T) acquisitions across various anatomical regions.

Conclusions:

• The FMFMS method offers superior performance in low signal-to-noise ratio (SNR) scenarios.
• Provides reliable water-fat separation across diverse field strengths and anatomical locations.