A prospective study comparing R2* derived Liver iron concentration(LIC) with noise-corrected post processing of data against FerriScan reported LIC in patients with liver iron overload.
Kartik Jhaveri1, Stephan Kannengiesser2, Nima Sadougi3, Marshall Sussman1, Hooman Hosseini-Nik3, Leila Zahedi3, and Richard Ward1

1UHN,University of Toronto, Toronto, ON, Canada, 2Siemens Germany, Erlangen, Germany, 3UHN, Toronto, ON, Canada


MRI is currently utilised as a non-invasive method for liver iron concentration (LIC) estimation and has essentially replaced liver biopsy. FerriScan derived LIC is considered the “gold standard” but has associated increased costs and delay results from required external data transmittal. There is no universal agreement or standardization of R2* derived LIC methods. We describe an optimized R2* method with noise corrected post processing of data for LIC estimation with simultaneous comparison to FerriScan derived LIC. Our results show very good correlation between R2* LIC and FerriScan LIC with potential for substitution of the latter with our R2* technique.


Patients with liver iron overload require frequent assessment of liver iron concentration (LIC) for therapy monitoring and complication risks ideally by non-invasive methods.FerriScan (Resonance Health, Claremont, WA, Australia is) is a well-established MRI technique for LIC estimation but adds costs and need for external data transmittal for procurement of results (1, 2). R2* based LIC estimations have been reported previously (3-5) however these are not standardised as the acquisition parameters,data post processing and conversion formulae are variable. Recently, Garbowski et al (6) have reported biopsy based calibration of a T2* method for LIC calculation and comparison to FerriScan.


To compare prospectively and intra-individually R2*derived LIC with noise-corrected post processing of data against FerriSscan reported LIC obtained on the same day.


127 patients (70 women, 57 men) with a mean age of 38 years (range, 19-72) prospectively underwent FerriScan and R2* MR imaging on the same day. Thalassemia was the most common diagnosis ( 82.7%). 80 %( 105/127) had history of blood transfusion, and 36% (48/127) had undergone splenectomy. Hemoglobin and ferritin levels were available in 118 patients, in which 90% (106/108) were anemic (Hb <120 g/l). The mean serum hemoglobin and ferritin were 101 g/L (range, 57-147 g/L) and 2040 ng/ml (range, 8-14094 ng/L) respectively. MRI was performed on all on 1.5T (MAGNETOM Aera, Siemens Healthcare, Erlangen,Germany).FerriScan acquisition using free breathing multi-spinecho sequence was as per mandated protocol. An optimised prototype 3D 6 echo breathhold gradient echo acquisition with lowest TE of ~1ms with TE range 1-9ms, constant TR 11.8ms was acquired immediately thereafter.(Fig.1)For each subject, the individual gradient-echo magnitude images were saved in DICOM format, and re-imported in Matlab (The MathWorks, Natick, MA, USA). Image pixels were fitted individually to a signal model containing R2* and optionally fat signal fraction (FF), as well as a Rician noise signal contribution as described in (7). A version of equation 3 (7) was used to predict the maximum likelihood (ML), i.e. the maximum of the probability density function, of the measured signal s. Two alternative signal models were used: $$s = ML(|m_0 \cdot exp(-R_2^* \cdot TE)| \quad \text{given} \, n \quad \text{(2a)}$$ and R2*-plus-FF (fat signal fraction)

$$s = ML(|m_0 \cdot ((1-FF) + c(TE) \cdot FF) \cdot exp(-R_2^* \cdot TE)| \quad \text{given} \, n \quad \text{(2b)}$$ where n is the noise level, and c(TE) are the complex fat signal dephasing factors. The function lsqcurvefit of Matlab R2012b was used for parameter fitting, with bounds of [0 inf] on all parameters, and otherwise standard settings, in particular using the "trust-region-reflective" algorithm. Manual ROIs were drawn in a homogeneous region of the right liver hemisphere. In a firstROI, each pixel was fitted individually to equation 2a and/or 2b with n as a free parameter. The average value of n across the ROI was taken as n0, excluding pixels for which the parameter fit failed (R2* = 0, or FF >= 50%). n=n0 as a fixed parameter was then used for fitting all pixels individually to equation 2, assuming a constant noise level across the liver. Inaccuracies in the value of n0 were deemed more acceptable than an increased parameter standard deviation potentially caused by the additional free model parameter.LIC values were derived from average R2* values in 3 additional ROIs drawn on post-processed data utilizing the equation published by Garbowski et al (6). LIC derived by R2* and FerriScan were compared using Pearson’s correlation, linear regression analysis and Bland-Altman plots.


Mean LIC, in mg Fe per g dry weight, derived by FerriScan was 9.7mg/g (Range 0.6-42, median 6.5), by R2* 9.2mg/g (range 0.9-36.9, median 6.8) and by R2*-plus- FF of 9.4(range 0.9-60.5, median 6.3). Comparing R2* -only LIC versus FerriScan, r2=0.9172, Pearson correlation 0.9577(95% CI: 0.9403, 0.9701) with Bland-Altman plots displaying mean difference =1.157(95% CI: -5.603, 7.906) were obtained(Fig.2,3). Comparing R2*plus-FF derived LIC against FerriScan LIC, r2=0.8423, Pearson correlation 0.9178(95% CI: 0.8763, 0.9457) with Bland-Altman plots displaying mean difference =1.283(95% CI: -7.507, 10.07) were obtained. Comparing R2* LIC against R2*-plus-FF LIC r2=0.9397, Pearson correlation 0.9694(95% CI: 0.9533, 0.9800) with Bland-Altman plots displaying mean difference =-0.0597(95% CI: -4.138, 4.018) were obtained.


R2*-derived LIC with pixel-wise, noise-corrected offline processing provides very good correlation with FerriScan-derived LIC. Compared with FerriScan, the fit-first followed by ROI-average of R2* strategy used in this work gives similar results and better r2 (>0.84) as the reverse order of operations and use of a truncation model published previously by Garbowski et al (r2=0.65).


Gerald R. Moran, Ph.D.Research Collaboration ManagerSiemens Healthcare Limited

Ravi Menezes, PhD, JDMI,University of Toronto.(Statistical Analysis)


1. St Pierre TG, Clark PR, Chua-anusorn W, Fleming AJ, Jeffrey GP, Olynyk JK,Pootrakul P, Robins E, Lindeman R. Noninvasive measurement and imaging of liver iron concentrations using proton magnetic resonance. Blood. 2005; 105:855–61.

2. http://www.resonancehealth.com/products/ferriscan-mri-measurement-of-liver-iron-concentration.html 3. Anderson LJ, Holden S, Davis B, Prescott E, Charrier CC, Bunce NH, Firmin DN, Wonke B, Porter J, Walker JM, Pennell DJ. Cardiovascular T2-star (T2*) magnetic resonance for the early diagnosis of myocardial iron overload. Eur Heart J. 2001; 22:2171–79.

4. Wood JC, Enriquez C, Ghugre N, Tyzka JM, Carson S, Nelson MD, Coates TD. MRI R2 and R2* mapping accurately estimates hepatic iron concentration in transfusion-dependent thalassemia and sickle cell disease patients. Blood. 2005; 106:1460–65.

5. Hankins JS, McCarville MB, Loeffler RB, Smeltzer MP, Onciu M, Hoffer FA, Li CS, Wang WC, Ware RE, Hillenbrand CM. R2* magnetic resonance imaging of the liver in patients with iron overload. Blood. 2009; 113:4853–55.

6.Garbowski MW, Carpenter JP, Smith G, Roughton M, Alam MH, He TG, Pennell DJ, Porter JB. Biopsy-based calibration of T2*magnetic resonance for estimation of liver iron concentration and comparison with R2 Ferriscan. J Cardiov Magn Reson 2014;16.

7.Feng Y, He T, Gatehouse PD, Li X, Harith Alam M, Pennell DJ, Chen W, Firmin DN. Improved MRI R2 * relaxometry of iron-loaded liver with noise correction. Magn Reson Med. 2013 Dec;70(6):1765-74.


Fig.1 R2* map of the liver obtained after noise corrected post processing of images acquired with an optimized prototype 3D 6 echo Gradient echo sequence.

Fig.2 Pearson correlation and linear regression analysis comparing Ferriscan LIC versus R2* LIC shows R2 = 0.9172, Slope = 0.7614 (95% CI: 0.7212, 0.8017), Pearson correlation: 0.9577(95% CI: 0.9403, 0.9701)

Fig.3. Bland-Altman plots of Ferriscan LIC versus R2* LIC shows Mean difference = 1.157 (95% CI: -5.603, 7.906)

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)