A fully automatic algorithm for assessing T2* and its certainty value for accurate cardiac and liver iron load determination

Abstract

Background: Quantification of myocardial and liver iron load has become a mainstay of guiding therapy in thalassaemia patients. However, current quantification methods are user dependent for data-point exclusion before curve fitting, and do not report the T2∗ certainty. Recently, an automatic inline maximum likelihood estimate (MLE) method with kspace Rician noise correction was validated against the reference standard manual truncation method. We now present a vendorindependent offline tool for fully automatic post processing and T2∗ quantification with certainty estimates, tested in computer simulations and patients with iron overload. Methods: A new hybrid method was implemented in Segment (http:// segment.heiberg.se), combining exponential fitting and residualweighting. The only user dependency is drawing the ROI for evaluation. Hybrid method: One of three fitting algorithms is applied based on an initial automated T2∗ estimation. If this estimate is shorter than the first echo time (TE1), an exponential model is used with offset correction and certainty residual-weighting; if the estimate is longer than 3×TE1, a pure exponential model residual-weighted to certainty of pixels is used; and if the estimate falls in-between TE1 and 3×TE1, a linear interpolation of the two methods is applied. Possible bias for large T2∗ values (>>maximum TE) is compensated for by using the magnitude of the fit residual error. Certainty estimates of T2∗ are calculated based on size of the fit residual error. For low T2∗ value certainty estimates the TEs are also taken into account. Simulations: Computer phantoms were generated for T2∗=0.5-40 ms with varying Rician noise. For each T2∗ and noise level 200 computer phantoms were created for calculation of T2∗ certainty. Patient study: Nine iron overload patients (5 male; median age 12, range 1.4-39 years) were scanned at 1.5T for cardiac and liver iron load using 2 multi-gradient echo sequences. All parameters were identical (voxel = 2 × 2 × 10 mm, matrix = 256, 10 echoes, FA = 20°, BW = 833 Hz, SENSE = 2) except for TE (cardiac: TE1/ΔTE 2.5/2.3 ms; liver: TE1/ΔTE 0.8/1.6 ms). The new hybrid method and the MLE method were applied in the cardiac septum and liver, respectively. Results: Results of computer phantom experiments are shown in Figure 1. In patients, bias ± SD between Segment and MLE were -0.28±0.41 ms (-1.16 ± 1.68%; Figure 2). Conclusions: Accurate iron-load T2∗ is provided by the proposed fully automatic method, comparable with the MLE method over a wide range. In addition to previous algorithms, the new hybrid algorithm reports a certainty estimate for T2∗. (Figure presented).