Comparison of liver iron concentration calculated from R2* at 1.5 T and 3 T

It is important that iron levels are estimated sufficiently accurately so that clinically significant changes in iron can be detected. LIC estimated from 1.5 T R2* is widely used and is established as a method which is regarded as having sufficiently reliable precision and accuracy for clinical practice [6].

However, 3 T-R2* LIC is not as widely used and there are fewer studies to verify its accuracy. In this study, 3 T-R2* LIC was compared with 1.5 T-R2* LIC, to assess whether 3 T-R2* LIC can be used by clinicians and researchers with the same confidence that 1.5 T-R2* LIC is used. Additionally, comparing 3 T-R2* LIC with 1.5 T-R2* LIC enables the assessment of whether 1.5 T-R2* LIC and 3 T-R2* LIC can be used interchangeably, for example when a patient undergoing monitoring is scanned at both 1.5 T and 3 T or if a clinical trial is performed at both field strengths. 30 participants were scanned on multiple scanners to enable this comparison.

One method of converting to 3 T R2* to LIC could be by converting 3 T R2* to 1.5 T R2* and then using a 1.5 T-R2* LIC calibration curve, In this study 1.5 T and 3 T LIC had a linear relationship which was as expected, as R2* is in theory directly proportional to field strength. It was similar regardless of whether all the data were included or whether it split by vendor and was similar to the relationship found by Alam, with 3 T R2* being about half of 1.5 T R2*. This is also similar to the relationship found in other studies [13, 14]. One way to obtain similar LIC across two field strengths therefore could be to firstly to covert 3 T R2* to 1.5 T R2* and then convert 1.5 T R2* to LIC using the preferred 1.5 T LIC calibration curve.

A direct 3 T-R2* LIC calibration curve is also available. In this study, the LIC found using d’Assignies’ 3 T-R2* LIC calibration curve was compared with Wood’s 1.5 T-R2* LIC and Garbowski’s 1.5 T-R2* LIC. The results showed there was strong correlation and bias of less than 0.2 mg/g dw for both 1.5 T-R2* LIC conversion methods. However lower limits of agreement of − 0.55 mg/g dw and − 0.32 mg/g dw were found when using Wood and Garbowski’s conversion, respectively. One reason for the difference between 1.5 T-R2* LIC and 3 T-R2* LIC may be due to differences in patient populations between 1.5 T and 3 T. In d’Assignies population there were 49 patients with iron levels of 0-2 mg/g dw [7], whereas in both Wood’s [2] and Garbowski’s [5] studies there was only one patient with LIC below 2 mg/g dw. This therefore could impact the resulting calibration curve.

Garbowski’s 1.5 T-R2* LIC had greater agreement with d’Assignies 3 T-R2* LIC compared to Wood’s 1.5 T-R2* LIC. Additionally, a smaller bias and tighter limits of agreement were found. This is probably due to there being greater similarities between d’Assignies’ study and Garbowski’s study compared to Wood. For example, the quantification of biopsies in d’Assignies’ study and Garbowski’s study was in the Rennes laboratory, whereas in Wood’s study quantification was performed at Mayo Medical Laboratory. Differences in methods in processing and analysing biopsies can yield different results, and so the same laboratory being used for d’Assignies’ and Garbowski’s study may be one reason for the similar results. Additionally, similarity between sequence methods may be a factor. Both d’Assignies and Garbowski used multi-echo gradient echo techniques, whereas Wood’s study used a single-echo acquisition, repeated multiple times. Wood’s study also contained fewer biopsies (22) [2] compared to Garbowski (50) [5] and d’Assignies (76 for generating the calibration curve used in this paper, 104 in total in the study) [7].

The calibration curves may also differ due to the use of different fitting algorithms and software to produce the curves, as highlighted by Meloni [15]. Wood used a model which included noise as a variable offset, Garbowski used a truncation model for noise and d’Assignies used a noise subtraction algorithm which subtracted mean background noise [2, 5, 7]. It should be noted that the method of noise subtraction used by d’Assignies may not be appropriate, given the Rician nature of MR noise [16].

From the Bland Altman plots there appears to be an increasing positive difference between the 3 T-R2* LIC and 1.5-R2* LIC when both Wood and Garbowski’s methods are used, which may indicate that at higher iron levels the difference between 3 T-R2* LIC and 1.5 T-R2* LIC may be even greater. However, further verification of this is needed by acquiring data in patients with higher LIC (as in this study all participants had R2* equivalent to LIC below 2.5 mg/g dw, regardless of method used). It may be that using d’Assignies’ equation which included patients with LIC above 7.26 mg/g dw would improve the agreement between the methods at higher LIC.

Comparison was also made when data were split into Siemens 2.89 T and Philips/GE 3.0 T. However, there was very little difference in the linear regression equations, and the correlation and agreement, when compared to the pooled data results. This is perhaps to be expected when using d’Assignies for 3 T-R2* LIC, as that study included both Siemens and Philips 3 T scanners [7].

The improved agreement, bias and limits of agreement between d’Assignies and Garbowski compared to d’Assignies and Wood indicate that careful choice is required when choosing calibration curves. Given that factors such as the population, acquisition technique and laboratory analysis may all influence the resulting calibration curve, it is important that, if using both 1.5 T and 3 T for estimating LIC within the same hospital/centre, the calibration curves used have been acquired using as similar methodology as possible. However, it should be noted that the bias when using Wood was still below 0.23 mg/g dw and so it still may be clinically acceptable to use Wood for 1.5 T-R2* LIC and d’Assignies for 3 T-R2* LIC. It should also be noted that it is preferable to use 1.5 T R2* for LIC evaluation in patients with high iron overload, as at 3 T the fast signal decay can make R2* estimation challenging.

There were several limitations to our study. Firstly, no biopsy paired LIC was available, which is currently accepted as the ground truth for LIC. However, 1.5 T-R2* LIC is widely accepted and so it would be hard to ethically justify performing biopsies when 1.5 T-R2* LIC is available. Secondly, this study only involved participants with R2* equivalent to LIC below 2.5 mg/g dw (regardless of R2* to LIC conversion). Given that mild iron overload is considered to be 3. 2 mg/g dw – 7 mg/g dw [10] no participants had iron overload and so are not representative of the majority of people being imaged to establish LIC levels. Further studies are required on participants with higher LICs, but this study will provide a good foundation whilst other studies are awaited. Thirdly, the study only contained 30 participants. However, this was enough to establish correlations and agreements between 1.5 T-R2* LIC and 3 T-R2* LIC. Another limitation of this work is that all data were acquired and analysed using the same techniques. Using different acquisitions technique, and different software, could impact the results.

In conclusion, this study suggests that LIC estimated using R2* measured at 3 T may be similar to LIC measured using R2* at 1.5 T. However, care should be taken on which method is used to convert from R2* to LIC.

留言 (0)

沒有登入
gif