An online open-source tool for automated quantification of liver and myocardial iron concentrations by T2* magnetic resonance imaging

Abstract

Objective:

To assess whether an online open-source tool would provide accurate calculations of T₂^* values for iron concentrations in the liver and heart compared with a standard reference software.

Methods:

An online open-source tool, written in pure HTML5/Javascript, was tested in 50 patients (age 26.0 ± 18.9 years, 46% males) who underwent T₂^* MRI of the liver and heart for iron overload assessment as part of their routine workup. Automated truncation correction was the default with optional manual adjustment provided if needed. The results were compared against a standard reference measurement using commercial software with manual truncation (CVI^42® v. 5.1; Circle Cardiovascular Imaging; Calgary, AB).

Results:

The mean liver T₂^* values calculated with the automated tool was 4.3 ms [95% confidence interval (CI) 3.1 to 5.5 ms] vs 4.26 ms using the reference software (95% CI 3.1 to 5.4 ms) without any significant differences (p = 0.71). In the liver, the mean difference was 0.036 ms (95% CI −0.1609 to 0.2329 ms) with a regression correlation coefficient of 0.97. For the heart, the automated T₂^* value was 26.0 ms (95% CI 22.9 to 29.0 ms) vs 25.3 ms (95% CI 22.3 to 28.3 ms), p = 0.28. The mean difference was 0.72 ms (95% CI 0.08191 to 1.3621 ms) with a correlation coefficient of 0.96.

Conclusion:

The automated online tool provides similar T₂^* values for the liver and myocardial iron concentrations as compared with a standard reference software.

Advances in knowledge:

The online program provides an open-source tool for the calculation of T₂^* values, incorporating an automated correction algorithm in a simple and easy-to-use interface.

INTRODUCTION

Iron quantification using T₂^* MRI has significantly modified the management of diseases with chronic iron overload.¹ Part of the limitation in widespread use of the technique occurs owing to restricted access to quantification software and difficulty in obtaining accurate numbers, especially in situations of severe iron overload where truncation or an offset model has to be applied.^2,3 In order to facilitate the calculation of T₂^* values in these situations, He et al⁴ published an accurate algorithm for automated truncation and correction of the analysis of T₂^* decay curves, simplifying the method while maintaining excellent accuracy with a coefficient of variation of only 1.6%. Despite the significant results, the technique was implemented only in MATLAB^® (MathWorks^®, Natick, MA), limiting its access in most clinical centres worldwide. Because of this limitation, we sought to develop an online open-source tool incorporating the described algorithm in order to promote wider availability of the automated process to researchers and clinical physicians without the need for dedicated software.

METHODS AND MATERIALS

In order to provide amplified access to the automated algorithm, a tool was written in pure HTML5/Javascript, incorporating the previously established rules while running in most modern web browsers (http://www.isodense.com/ic) (Figure 1). To test the tool, we selected 50 patients who underwent T₂^* MRI of the liver and heart for iron overload assessment as part of their routine work-up (mean age 26.0 ± 18.9 years, 46% males, 80% with thalassaemia major as the primary haematological disorder). T₂^* acquisition was performed after patients completed the institutional review board-approved informed consent in a 1.5-T scanner (MAGNETOM^® Aera; Siemens, Erlangen, Germany). A short-axis single-slice black-blood image of the heart in the mid-ventricular level as well as an axial slice of the liver using multiecho gradient-echo images was obtained according to previous published methods.^5,6 For each image, we identified the individual echo times (TEs) and region-of-interest-based signal intensities for each organ in the scanner console itself. The operator would draw the region of interest on each T₂^* image and then manually insert these values in the online webpage. Automated truncation was set as the default, but the user could manually adjust the results if truncation failed (for example, in cases with fewer than three data sets). For comparison against a standard reference, the same data sets were loaded in a commercial software (CVI^42® v. 5.1; Circle Cardiovascular Imaging) and regions of interest drawn as close as possible to the original drawing on the console in the T₂^* module of the software with manual truncation for correction.²

Figure 1.

A sample calculation performed with the online tool that allows for automated truncation and provides the results in T₂^*, R2* and final liver and myocardial iron concentrations. LIC, liver iron concentration; SI, signal intensity; TE, echo time.

Values obtained with both tools were compared using paired t-test, Bland–Altman plots and correlation coefficients (MedCalc v. 15.2; MedCalc Software, Ostend, Belgium).

RESULTS

14 patients (28%) had myocardial T₂^* values below the normal 20 ms cut-off while only 2 patients (4%) had normal liver iron levels with 22 patients (44%) with severe iron overload. The mean myocardial T₂^* values calculated by the reference software was 25.3 ms [95% confidence interval (CI) 22.3 to 28.3 ms] corresponding to a myocardial iron concentration of 0.87 mg g⁻¹ (95% CI 0.76 to 1.02 mg g⁻¹).⁷ No significant differences were observed comparing these values to the myocardial T₂^* values calculated using the online tool of 26.0 ms (95% CI 22.9 to 29.0 ms; p = 0.28). For the liver, the mean T₂^* value measured using the reference software was 4.26 ms (95% CI 3.1 to 5.4 ms) corresponding to a liver iron concentration of 7.4 mg g⁻¹ (95% CI 5.7 to 10.1 mg g⁻¹).⁸ No significant differences were observed compared with the online calculation of 4.30 ms (95% CI 3.1 to 5.5 ms; p = 0.71).

The results for the liver and heart demonstrated excellent agreement with the reference method (Figure 2). In the liver, the mean difference was 0.036 ms (95% CI −0.1609 to 0.2329 ms) with a regression correlation coefficient of 0.97; in the heart, the mean difference was 0.72 ms (95% CI 0.08191 to 1.3621 ms) with a regression correlation coefficient of 0.96.

Figure 2.

Bland–Altman plots and scatter diagrams with regression line demonstrating the comparison of T₂^* values in the liver (a, b) and heart (c, d) using the online tool vs the standard offline calculations. SD, standard deviation.

DISCUSSION

One of the main limitations to the implementation of routine iron overload assessment in many centres in the world is the inability to accurately calculate T₂^* values with currently available tools. Not only that, training and understanding the main limitations of these calculations are especially important in order to avoid making significant mistakes in the final iron values by underestimating the true iron concentrations in cases where severe iron overload exists.⁹ Although some authors suggest that using an offset model might solve part of the problems with noise in the current gradient-echo sequences, the true T₂^* decay curve appears to be best reflected by a monoexponential formula,¹⁰ especially when analysing the data using region-of-interest-based methods as is most commonly performed by routine commercial software.¹¹ While CVI42 provides an option to calculate the T₂^* values using baseline correction, we chose not to include this function in the online tool so as not to add extra variability to the results provided, especially considering the possible use of this tool in less experienced centres. While we recognize that the choice of methods merits scientific discussion, in clinical practice it appears that the differences are not specifically appreciable and so we opted for a more simple unique approach.¹²

With the monoexponential model, the application of truncation of the last data points is fundamental in order to obtain accurate results in cases with severe iron concentration. Lack of application of this principle accounts for most of the errors in the interpretation of the examination in less experienced centres.³ Therefore, the development of an automated algorithm by He et al⁴ reducing the need for user manipulation of the data with correction for imperfections in the original T₂^* signal was greatly sought. This algorithm incorporates the following rules in order to correct for the apparent offset in the decay curve: if the original correlation coefficient is >0.995 for all data sets, it accepts the original T₂^*; if not, it automatically eliminates the last data points until the correlation coefficient exceeds that threshold; the elimination proceeds until the T₂^* values drop to <2.5%. While the method is very effective in providing corrected numbers, the need to use the original implementation in MATLAB significantly limits its availability, as most clinical centres do not routinely use that software. With the incorporation of the same rules previously described in an online platform with an easy-to-use interface and graphical visualization of the original fitting and corrected curves, we believe that more centres might be able to perform the accurate interpretation of T₂^* examinations.

While the algorithm corrects for most cases of noise in the images, in liver cases with very severe iron overload (T₂^* < 1.0 ms) or when the initial TE is higher than 1.3 ms, the automated truncation cannot always be performed. In these cases, users will have to choose to use a maximum of three or even two data points manually, increasing the overall spread of CIs of the calculated iron concentrations. While this can become a limitation, it has more to do with the restraints of the original multiecho gradient-echo technique than with the software used to analyse the data per se.³ Added to this limitation, the results presented in this article were produced in a single-centre setting with experienced users in T₂^* analysis and there is uncertainty as to how the online tool might perform in less proficient centres where other difficulties might appear. Finally, while one might expect the exact same results in both platforms, as the exponential equation is the same, slightly different regions of interest used for each software may explain part of the variation observed. This can also be explained by the fact that many of the patients studied presented with only mildly abnormal values specifically in the heart where other factors besides iron concentration affect the accuracy of the measurements at such levels.¹³

CONCLUSION

In conclusion, the online open-source tool provides accurate T₂^* values, incorporating a previously published automated algorithm in an accessible platform for clinical centres worldwide without the need for additional software while correcting for the most common source of error in the interpretation of these examinations.

FUNDING

This work was supported by the Fundacao de Amparo a Pesquisa do Estado de São Paulo (FAPESP)—Public Agency and Sultan Bin Khalifa International Thalassemia Award.