There are several publications concerning liver relaxometry based on fitting of magnitude MRI data. However, the full information is contained in complex data (1). This work was performed to compare both approaches directly in identical ROIs on real patient data.94 regular transfused patients suspected for liver iron overload were investigated with multi-contrast GRE-MRI at 1.5 T (Siemens Avanto) using multiple T_E/T_R/FA, cf. Table 1, obtaining five transversal slices positioned over the liver. Two slices best suited for comprising vessel-free parts of liver tissue were analyzed by placing manually ROIs of fixed size containing 45 voxels, three in each slice. R₂* relaxometry of averaged ROI signal values was performed, as conjoined fit of all FA and both echo-spacings, in two manners: a) using only magnitude images, modelling noise as free fit parameter with the expected signal as sum of squares of noise and ideal signal, and b) involving complex information (1). For both fitting processes, performed at identical ROIs, modulation of GRE signal caused by fat/water-dephasing was considered according to (1). Not only R₂* values, but also their uncertainties were determined by the fit algorithm. Examinations were excluded from further analysis if fit failed to converge in more than three of six ROIs. Median R₂* values for ROIs were determined for each patient seperately for both methods. Correlation between results was studied, and mean relative uncertainties for all patients and both methods were calculated.While magnitude fit worked in all cases, three examinations (3.1%) had to be excluded since the complex fit did not converge sufficiently. Correlation and correspondence of R₂* values was good, cf. Fig. 1, yielding an R² value of 0.985. Slope was calculated as 0.937, indicating that magnitude fit returned slightly larger values in some cases. The mean relative uncertainty of R₂* values was 2.4 ± 1.8 % for magnitude and 12.8 ± 9.4 % for complex fit. Uncertainties as a function of resulting R₂* values for both approaches are shown in Figs. 2 and 3.

Multipeak fat-corrected R₂* relaxometry worked on data acquired in a number of patients. ROI based analysis was employed to reduce uncertainty of measured data, assuming this would also diminishing result uncertainy. For three examinations, complex fit failed to converge, probably due to phase inconsistencies within ROIs, whereas magnitude fit worked well on all patients.

R₂* values are comparable, however, slightly larger in some cases with the magnitude approach, yielding a tendency similar to that demonstrated in (1).

Uncertainty was lower for the magnitude fit by about a factor of five. This is surprising at the first sight since not magnitude, but complex fit considers complete information contained in MRI data. However, larger uncertainties for complex fit may again be caused by the ROI based approach. Inconsistent phase evolution, i.e. dephasing between different voxels in one ROI, leads to signal reduction for the averaged signal and may cause observed uncertainties.

Multiple breathholds with different FA were performed to get sufficient data for reliable fits in patients showing large R₂*, which were present in our cohort of regular transfused patientes. These multiple acquisitions improve magnitude fit, but may impair complex fit because of scanner phase instabilities. To evaluate influence of phase incongruities within ROIs, voxel-wise fit should be studied. Though, in our preliminary experience, single-voxel fits frequently failed to converge especially for large R₂* and showed large uncertainty of results. Both findings are probably due to larger noise in one voxel than in averaged ROI data. Furthermore, single-voxel fit is time-consuming and therefore less suited for routine application.

In our implementation of ROI based analysis, fit of magnitude data yielded R₂* values comparable to fit involving complex data, at improved fit stability and reduced R₂* uncertainty.