Synopsis

Keywords: Fat, MSK

Fat signals can obscure morphological structures and influence quantitative parameter mapping in ultrashort echo time (UTE) musculoskeletal MRI. However, conventional fat suppression methods have challenges in UTE imaging of short T₂ species. For example, the chemical-shift based fat saturation can significantly attenuate short T₂ signals with broad spectra, and water excitation with long composite pulses may yield a long minimum echo time (TE). Alternatively, the feasibility of single-point Dixon (1p-Dixon) has been demonstrated for UTE imaging. In this study, we investigate the feasibility of 1p-Dixon based on the 3D phase modeling approach, which achieves fat suppression without additional data acquisition.

Introduction

Strong signal from fat is one of the major confounding factors obscuring morphological structures and complicating quantitative parameter mapping in musculoskeletal (MSK) MRI¹. Many fat suppression techniques have been developed for clinical MRI, including fat-saturation^2-6, inversion recovery-based methods³, m-point Dixon^4,5, and water excitation⁶. Recently, ultrashort echo time (UTE) imaging has been extensively investigated for MSK imaging. However, conventional fat suppression methods have challenges in UTE imaging of short T₂ species^7,8. The m-point Dixon is one of the feasible techniques for fat suppression in UTE imaging. However, multiple images are required to model the fat and water signal, which imposes a longer scan time. Recently, the feasibility of single-point Dixon (1p-Dixon) in UTE-MSK imaging has been demonstrated^8,9. The previously proposed 1p-Dixon UTE imaging requires additional image acquisition to compensate for B₁ and B₀ inhomogeneity⁸ or a complicated process with manual parameter tuning¹⁰. In this study, we present a simple but robust approach to achieve 1p-Dixon using 3D phase modeling.

Methods

1p-Dixon decomposes the water signal from the fat signal using the measured complex MRI signal, $$$S_{exp}$$$ that can be modeled as:

$$S_{exp} = (W + Fe^{i\theta(t)})e^{i\phi}\qquad(1)$$

where t is the time delay in free induction decay, W and F are the magnitudes of the water and fat signals, $$$\theta$$$ denotes the phase evolution due to chemical shift of fat, and $$$\phi$$$ denotes additional phase error due to B₁ and B₀ field inhomogeneity. If the phase error, $$$\phi$$$, is removed, the complex signal can be directly decomposed to W and F based on the fat signal model:

$$\frac{Im(s)}{sin(\theta(t))}\qquad(2)$$
$$W=Re(S)-Fcos(\theta(t))\qquad(3)$$

where Im and Re are the operators to take imaginary and real parts in the complex signal. $$$\phi$$$ is slowly varying in space and can be shown to roughly resemble a low-order polynomial from Maxwell’s equation. In the proposed approach, the phase map was first unwrapped and then fitted to a 3D phase model based on the 4th-order polynomials represented in Equation 4:

$$ \phi(x,y,z)= \sum_{k=0}^{3}\sum_{j=0}^{3-k}\sum_{i=0}^{3-j-k}C_{ijk}x^iy^jz^k + \sum_{n=0} C_{n4}x^iy^{4-i}\qquad (4)$$
$$C =argmin_c(\sum_{i,j,k=1}^{n}(<S_{exp} - \phi(C,x_i,y_j,z_k))^2)\qquad(5)$$

where C represents all coefficients to be minimized. The baseline phase for water and $$$\theta$$$ for fat were approximated from the distribution of the corrected phase after removing $$$\phi$$$. The outliers with phase values two standard deviations away from the mean were cropped. Then, the refined phase values were shifted to ensure all values are positive, and the maximum phase was taken as $$$\theta$$$.

To validate the proposed approach, MRI experiments were conducted using a 3D UTE-Cones sequence implemented in a 3T clinical MRI scanner (MR750, GE Healthcare) as shown in Figure 1A. An experiment was performed with a phantom comprised of two sets of tubes made of lard and peanut oil respectively, where each set included six tubes with different fat fractions (0%, 10%, 20%, 40%, 60%, and 100%). An ex vivo experiment was conducted with four cadaveric knee joints. Finally, an in vivo experiment was conducted with two healthy volunteers. The detailed imaging parameters are shown in Figure 1B.

All data processing was performed in Matlab. The raw data from individual coils were reconstructed using the NuFFT¹¹ and were combined into one complex image inputted to the proposed 1p-Dixon pipeline. The least squares minimization shown in Equation 5 was solved using the Levenberg–Marquardt algorithm.

Results

Figure 2 shows an example demonstrating the fitted 3D phase model of $$$\phi$$$ and the line profile, as well as the resultant $$$\phi$$$-corrected phase image. The smoothly varying phase error was dramatically suppressed in Figure 2D by using the proposed phase modeling approach, as also shown with the line profiles in Figure 2B.

Figure 3 shows the results from the phantom experiment. After correction of $$$\phi$$$, 1p-Dixon yielded reliable fat and water separation. The estimated fat fraction showed high linearity with respect to the actual fat fraction (Figure 3H), with Pearson’s correlation (R) > 0.96 and p-value < 0.05.

Figure 4 shows representative slices from all four cadaveric knee specimens, where robust fat suppression was achieved without negative impacts on the signal from short T₂ tissues. Figure 5 shows the results of two healthy volunteers. In the estimated water images, short and long T₂ tissues of interest are clearly delineated, including tendons (yellow arrows), ligaments (red arrows), muscles (green arrows), and cartilages (blue arrows).

Discussion and Conclusion

The proposed 1p-Dixon approach utilizing polynomial fitting of the 3D phase model is likely to provide robust and reproducible fat suppression for UTE imaging without the need for additional data acquisition. The underlying assumption for lower-order polynomial modeling of is highly plausible for UTE imaging where B₀ inhomogeneity-induced phase error is negligibly small and thus smooth. The B₁ phase, which depends on the electric conductivity of tissues and the receiver phase of the RF coil¹², is known to be smooth. We used a data-driven approach based on a histogram of phase values to find the baseline phase (water phase) and determine the best fat phase. Although we demonstrate this approach is effective in morphological UTE-MSK imaging, this may cause a potential error in quantitative UTE imaging. In future studies, we will further improve the proposed method, and apply it to quantitative UTE imaging.

Acknowledgements

The authors acknowledge grant support from the NIH (R01AR078877, R01AR062581, R01AR068987, and RF1AG075717), VA Clinical Science Research and Development Services Merit Awards (I01CX002211), the DFG (SE 3272/1-1), and GE Healthcare.