Phase-sensitive modelling improves Fat DESPOT multiparametric relaxation mapping in fat-water mixtures

TL;DR

This work demonstrates the advantages of the GC approach for initial guesses paired with complex fitting for Fat DESPOT multiparametric imaging.

Abstract

Purpose: To improve on the original form of Fat DESPOT, a multiparametric mapping technique that returns the fat- and water-specific estimates of $R_1$ ($R_{1f}$, $R_{1w}$), $R_2^*$ , and proton density fat fraction (PDFF) by upgrading the fat-water separation method used for selection of initial parameter guesses, and by introducing explicit model sensitivity to the phase of the water and fat signals. Methods: We compared the 3-point Dixon and Graph Cut (GC) approaches to initial guesses for Fat DESPOT in phantom experiments at 3 T in a variable fat fraction gel phantom. Also in phantom, we then compared the original Fat DESPOT approach to a magnitude approach modelling the phases of fat and water separately (Fat DESPOT$_{mφ}$), and an approach that models the complex data (Fat DESPOT$_c$). The best-performing approach was then used in the lower leg of a healthy human participant. Results: In phantoms, Fat DESPOT using the 3-point Dixon and GC performed similarly in parametric estimates and precision, though the Dixon approach deviated from the overall trend in the 50% nominal fat fraction ROI. Furthermore, Fat DESPOT$_c$ showed the best agreement with reference PDFF (average error 1.5 +/- 1.2%) and the lowest combined standard deviation across ROIs, for PDFF, $R_{1f}$, and $R_{1w}$ (σ = 0.13%, 0.19 s$^{-1}$, 0.0082 s$^{-1}$). Conclusion: With a higher precision of $R_{1f}$ and $R_{1w}$ , accuracy of PDFF, and more echo time versatility than other compared approaches, this work demonstrates the advantages of the GC approach for initial guesses paired with complex fitting for Fat DESPOT multiparametric imaging.

Figures (10)

• Figure 1: Regions of interest of (a) the variable fat fraction phantom and (b) the lower leg of a human volunteer. In the phantom, ROIs 1-7 correspond to nominal fat fractions of 0%, 5%, 25%, 50%, 60%, 75%, and 100% respectively. In the lower leg, ROIs 1-3 correspond to tubular bone marrow, calf skeletal muscle, and subcutaneous fat. Bone marrow and subcutaneous fat voxels of interest within the ROI were selected based on a PDFF estimate $>$70% and $>60$ respectively. All ROIs were measured over 3 slices of the acquired image.
• Figure 2: Multiparametric maps for PDFF, $R_2^*$, $R_{1f}$, $R_{1w}$, and nRMSE using the 3-point Dixon and GC as PDFF initial guess inputs for Fat DESPOT$_m$ on a 2$\times$6-echo dataset. To reduce noise in the $R_{1f}$ images, voxels with PDFF$<$3% and in the $R_{1w}$ images, voxels PDFF$>$97% were masked.
• Figure 3: Distribution of voxel-wise estimates of PDFF, $R_2^*$, $R_{1f}$, and $R_{1w}$, using the 3-point Dixon and GC as PDFF initial guess inputs for Fat DESPOT$_m$ on a 2$\times$6-echo dataset. Box = interquartile range, horizontal line = median, feathers= 1st and 4th quartile, dots= outliers.
• Figure 4: Multiparametric maps for PDFF, $R_2^*$, $R_{1f}$, $R_{1w}$, and nRMSE using Fat DESPOT$_m$, Fat DESPOT$_{m\phi}$, and Fat DESPOT$_c$ on an 8-echo dataset. To reduce noise in the $R_{1f}$ images, voxels with PDFF $<$ 3% and in the $R_{1w}$ images, voxels PDFF $>$ 97% were masked. Voxels outside the phantom were masked.
• Figure 5: Distribution of voxel-wise estimates of PDFF, $R_2^*$, $R_{1f}$, and $R_{1w}$, using Fat DESPOT$_m$, Fat DESPOT$_{m\phi}$, and Fat DESPOT$_c$ on an 8-echo dataset. Box = interquartile range, horizontal line = median, feathers= data range, dots= outliers.