Original Research

MR Characterization of Hepatic Storage Iron inTransfusional Iron Overload

Haiying Tang, PhD,1* Jens H. Jensen, PhD,2 Christina L. Sammet, PhD,3 Sujit Sheth, MD,4

Srirama V. Swaminathan, PhD,5 Kristi Hultman, PhD,6 Daniel Kim, PhD,7 Ed X. Wu, PhD,8

Truman R. Brown, PhD,2 and Gary M. Brittenham, MD4

Purpose: To quantify the two principal forms of hepaticstorage iron, diffuse, soluble iron (primarily ferritin), andaggregated, insoluble iron (primarily hemosiderin) using anew MRI method in patients with transfusional ironoverload.

Materials and Methods: Six healthy volunteers and 20patients with transfusion-dependent thalassemia syn-dromes and iron overload were examined. Ferritin- andhemosiderin-like iron were determined based on the mea-surement of two distinct relaxation parameters: the“reduced” transverse relaxation rate, RR2, and the“aggregation index,” A, using three sets of Carr-Purcell-Meiboom-Gill (CPMG) datasets with different interechospacings. Agarose phantoms, simulating the relaxationand susceptibility properties of tissue with different con-centrations of dispersed (ferritin-like) and aggregated (he-mosiderin-like) iron, were used for validation.

Results: Both phantom and in vivo human data con-firmed that transverse relaxation components associatedwith the dispersed and aggregated iron could be sepa-rated using the two-parameter (RR2, A) method. The MRI-determined total hepatic storage iron was highly corre-lated (r¼0.95) with measurements derived from biopsy or

biosusceptometry. As total hepatic storage iron increased,the proportion stored as aggregated iron became greater.

Conclusion: This method provides a new means fornoninvasive MRI determination of the partition of hepaticstorage iron between ferritin and hemosiderin in ironoverload disorders.

Key Words: iron overload; hepatic storage iron; ferritin,hemosiderin, iron quantification; MRIJ. Magn. Reson. Imaging 2014;39:307–316.VC 2013 Wiley Periodicals, Inc.

IN PATIENTS WITH iron overload, the amount of ironin functional and transport pools changes onlyslightly (1). Almost all of the excess is sequestered instorage forms, as diffuse, soluble, and rapidly mobiliz-able ferritin iron, and as aggregated, insoluble andslowly exchangeable hemosiderin iron. Ferritin iron isfound in virtually all cells, providing both an accessi-ble reserve of iron for synthesis of functional iron-con-taining compounds and a means of sequestering ironin a soluble, relatively nontoxic form within the cyto-sol (2). The ferritin iron core is composed of thehydrous ferric oxide mineral ferrihydrite with the ap-proximate formula 5Fe2O3�9H2O (3). Recent evidenceindicates that iron incorporation and release from fer-ritin are intrinsic, autonomous properties of the mole-cule, based on an equilibrium with the concentrationof cytosolic low-molecular-weight iron (4). Hemosid-erin, formed within secondary lysosomes (sidero-somes) from agglomeration of iron cores derived fromdenatured ferritin (5,6), seems to help protect againstiron toxicity by storing excess iron away from thecytosol (7). The scarce data available suggest that theamounts and distribution of these two forms of stor-age iron are influenced by the underlying disorder(such as hereditary hemochromatosis, thalassemia,sickle-cell disease, and other refractory anemias), bythe duration and extent of iron loading, and by thetype of therapy (phlebotomy, iron chelation) (8–10).Information is limited, in part, because the sole estab-lished means of separately measuring ferritin and he-mosiderin iron in tissue requires biochemical analysis

1Imaging, Discovery Medicine & Clinical Pharmacology, Bristol MyersSquibb, Princeton, New Jersey, USA.2Radiology and Radiological Science, Medical University of SouthCarolina, Charleston, South Carolina, USA.3Radiology, Northwestern University, Chicago, Illinois, USA.4Pediatrics, Columbia University, New York, New York, USA.5Philips Healthcare, Andover, Massachusetts, USA.6Mayo Clinic, Rochester, Minnesota, USA.7Radiology, The University of Utah, Salt Lake City, Utah, USA.8Electrical and Electronic Engineering, The University of Hong Kong,Hong Kong.

Contract grant sponsor: National Institutes of Health; Contract grantnumbers: R01-DK069373, R01-DK066251, R01-DK049108.

Contract grant sponsor: American Heart Association; Contract grantnumber: 0730143N.

Contract grant sponsor: Hong Kong Research Grant Council; Con-tract grant number: GRF HKU 7794/07M.

Contract grant sponsor: St. Giles Comprehensive Sickle-Cell – Tha-lassemia Program.

*Address reprint requests to: H.T., Bristol Myers Squibb, E1-207,Route 206 and Province Line Road, Princeton, NJ 08543. E-mail:haiying.tang@bms.com

Received July 16, 2012; Accepted March 15, 2013.

DOI 10.1002/jmri.24171View this article online at wileyonlinelibrary.com.

JOURNAL OF MAGNETIC RESONANCE IMAGING 39:307–316 (2014)

VC 2013 Wiley Periodicals, Inc. 307

of specimens obtained by biopsy. Clinically, only thetotal storage (ferritin plus hemosiderin) iron concen-tration of tissues is typically determined. Separatemeasurements of the two iron storage forms are rarelyreported, in part because of the limited tissue samplesobtained at biopsy. Progress in understanding themolecular mechanisms underlying iron overload (10)suggests that separate measurement of ferritin andhemosiderin iron would improve our understanding ofthe pathophysiology of iron-induced tissue damage.Clinically, separate measurements potentially couldprovide an early indication of iron toxicity and be use-ful in monitoring treatment by phlebotomy or with dif-ferent iron-chelating agents (11).

Storage iron within tissues influences the magneticresonance signal by altering the local magnetic fieldand strongly alters the signal intensity in both trans-verse relaxation time (T2) and effective transverse relax-ation time (T2*) weighted images (12–28). Ferritin andhemosiderin iron influence MRI signal decay throughdifferent mechanisms. At present, MRI methods esti-mate total storage iron by determination of either trans-verse relaxation rate (R2) or effective transverserelaxatoin rate (R2*); no information about the partitionof storage iron between ferritin and hemosiderin is pro-vided. The relationship between R2* and the total stor-age iron concentration has been considered to be linear(27). By contrast, the relationship between R2 and thetotal storage iron concentration is curvilinear over aclinically relevant range of iron concentrations (24,27),a phenomenon that is incompletely understood. An al-ternative approach to the MR characterization of tissueiron has been developed, based on a theoretical model(29), that uses quantitative determinations of the con-centrations of dispersed and aggregated iron to mea-sure the total storage iron. In brief, the differences inaggregation and solubility that permit physical separa-tion of ferritin and hemosiderin iron for biochemicalmeasurements also result in distinct MRI signal decaybehaviors (29). The signal decay caused by dispersed,soluble ferritin iron has a monoexponential form, andR2 can be derived from monoexponential fitting to amultiple spin echo (MSE) signal. In contrast, the signaldecay caused by aggregated, insoluble hemosideriniron is nonmonoexponential and strongly dependentupon the interecho time. Fitting the model to the multi-ple sets of MSE data with different interecho spacingspermits calculation of a “reduced relaxation rate” (RR2)and an “aggregation index” (A). The ferritin and hemo-siderin iron concentrations can then be estimated fromRR2 and A, respectively, and the total storage iron cal-culated as their sum.

Validation in vitro of this model has been reported inagarose phantoms that simulate dispersed (ferritin-like)iron with manganese chloride and aggregated (hemo-siderin-like) iron with iron oxide microspheres, togetherwith examination of a small number of subjects todemonstrate the feasibility of human studies (30). Thepresent study focuses on separate measurements ofdispersed (ferritin-like) and aggregated (hemosiderin-like) iron in patients with transfusion-dependent tha-lassemia syndromes and in healthy subjects. We dem-onstrate that the contributions of dispersed and

aggregated iron to the transverse relaxation rate (R2)can be separated using the parameters RR2 and A.Finally, the influnce of the stimulated echo (31) causedby the present MSE sequence on RR2 and A (32,33),and on iron measurement are analyzed based on phan-tom simulations. We hypothesize that tissue iron quan-tification based on the two parameters, RR2 and A,rather than the single parameter, R2 or R2*, willimprove estimates of total storage iron. We suggest thatseparate measurements of dispersed and aggregatediron may contribute to an improved understanding ofthe pathophysiology of iron overload, potentially provideearly warning of an increased risk of iron-induced tox-icity and permit rapid monitoring of treatment by phle-botomy or administration of iron-chelating agents.

MATERIALS AND METHODS

Theory

The MSE signal decay caused by ferritin iron has amonoexponential form, where R2 may be derived frommonoexponential fitting of a MSE signal. The R2

depends linearly both on the iron concentration andthe applied field (34). In contrast, because hemosid-erin iron is aggregated in insoluble clusters, theMSE signal decay caused by aggregated iron is notmonoexponential and has a strong dependence on theinterecho time (29). More precisely, the nonmonoexpo-nential decay is described approximately by the fol-lowing analytic form:

S ¼ S0 � exp �RR 2 � TEð Þ� exp �A3=4 Dtð Þ3=4 TE � tsð Þ3=8

� �: (1)

The time shift, ts, is defined to be:

ts ¼ 2t 1� t=Dtð Þ2h i

; (2)

where S0 is the initial signal intensity, spin echoesform at echo time (TE)¼2t, 2tþ2Dt, 2tþ4Dt,2tþ6Dt, etc., A is the aggregation index, and RR2 isthe “reduced” transverse relaxation rate. The parame-ter A, accounting for the nonmonoexponential term, isprimarily sensitive to hemosiderin iron, while the ex-ponential factor RR2, incorporating the reduction inthe measured R2, is primarily sensitive to the ferritiniron (29). It is advantageous to set t to be less than Dtto better sample the initial part of the decay curve,especially when the iron concentration is high.Assuming that the aggregated (hemosiderin-like) ironconcentration, CA, is proportional to the aggregationindex A, and the dispersed (ferritin-like) iron concen-tration, CD, is linearly related to the ferritin iron, thetotal iron concentration may be estimated as:

CT ¼ CD þ CA; (3)

where

CD ¼ a1 þ a2 � RR 2; (4)

308 Tang et al.

CA ¼ a3 � A; (5)

and a1, a2, a3, are empirical calibration parameters,which can be determined from a best fit of the MRImeasurements to independent estimates of the totalstorage iron concentration derived from biochemicalanalysis of biopsy specimens (35) or from biosuscep-tometry (1) using superconducting quantum interfer-ence device (SQUID). A unique feature of thisapproach is that it yields separate estimates for theferritin-like (CD) and hemosiderin-like (CA) ironconcentrations.

Human Subjects

Six healthy volunteers and 20 patients with transfu-sion-dependent thalassemia syndromes and ironoverload (19 to 51 years of age, 10 males, and 10females) were examined in a manner consistent withthe Institutional Review Board policies of our institu-tion. Written informed consent was obtained fromeach subject. The hepatic storage iron concentrationsof the patients were monitored by either biosuscep-tometry (n¼12), on the same day as the MRI meas-urements, or by biopsy (n¼8, median intervalbetween MRI and biopsy, 6 months; range, 1–11months). Because measurements by biosusceptome-try in vivo and biochemical analysis in vitro of hepatictissue in biopsy samples are closely correlated(r¼0.98; P<10�5) and do not differ significantly (36),the results of the two methods are usedinterchangeably.

Pulse Sequence for In Vivo MRI

All MR measurements were performed using a Philips1.5 Tesla (T) whole body MR scanner (Philips MedicalSystems, Best, The Netherlands) equipped with a five-element phased array cardiac coil. The scanner had amaximum gradient strength of 33 mT/m and a slewrate of 100 T/m/s. For each subject, three slices of10 mm thickness were acquired with a matrix of 128� 128 within a field of view (FOV) of 37 � 37 cm2 andwith an interslice gap of 1 mm. Each slice wasacquired in a double oblique orientation to includeboth the heart and liver.

To estimate RR2 and A, three CPMG-based MSEacquisitions were performed using different interechotimes, with the first echo at 4 ms (2s) for all sequen-ces. The imaging protocol includes: a 25-echosequence with an interecho time (2Dt) of 4 ms, a 15-echo sequence with an interecho time of 8 ms (i.e.,subsequent to the first 4-ms echo), and a 10-echosequence with an interecho time of 15 ms, each withECG triggering and respiratory navigator gating tominimize cardiac and respiratory motion artifacts.The repetition time (TR) was one heart beat, the paral-lel imaging SENSE factor was 1.5, the excitation pulseflip angle was 90

�, and the refocusing pulse flip angle

is 160�, the vendor’s default MSE implementation.

The image acquisition time was approximately 176 sfor each sequence.

In addition, a 17-echo multiple gradient echo (MGE)sequence was implemented for R2* measurement,with the first echo time occurring at 2.9 ms followingthe initial radiofrequency (RF) excitation, an interechotime of 0.83 ms, TR¼one heart beat, and excitationpulse flip angle¼50

�. The image acquisition time was

53 s. The MGE acquisitions were performed with ECGtriggering and respiratory navigator gating.

Image Analysis

Parametric maps for A and RR2 were calculated. Forhepatic iron quantification, a region of interest (ROI)was delineated to cover the majority of liver tissuewithin an image slice, and was propagated along theecho dimension. The liver tissue was further sepa-rated from the blood vessels and signal void regionsusing a threshold-based segmentation, to avoid signalcontributions from these regions. The three MSEdecay data within the liver tissue were rescaled, sothat the mean intensities at t¼4 ms were identical, tocorrect for inter-scan differences in intensity scaling.The corrections were typically a few percent of themaximum signal.

The two parameters, RR2 and A, were estimated byglobal fitting of the three MSE decay data to the non-monoexponential decay model based on Eq. [(1)]. Theconventional R2 was obtained using the MSE datawith the interecho time 4 ms, and R2* was obtainedfrom the MGE data, by fitting these signal decay datato the mono-exponential model. To reduce the effectof background signal on the parameter estimates, theactual functional expression consistently fit to allMSE and MGE data had the form:

Sfit¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiS2

ideal þ s2

q; (6)

where Sideal is the ideal (or unbiased) signal and r isthe mean background noise (30,37). The fits wereobtained by using the standard Levenberg-Marquardtmethod of least squares minimization, with four freeparameters (S0, RR2, A, and r) derived from the two-parameter model fits, and with three free parameters(S0, R2, or R2*, and r) derived from the mono-expo-nential (single-parameter) model fits.

The calibration parameters a1, a2, and a3 weredetermined from a linear least squares fit of Eq. [(3)]to the independent total iron estimates (CT) derivedfrom the biopsy or biosusceptometry measurements.With this calibration, MRI estimates for the totalCT ;MRI

� �, dispersed CD;MRI

� �, and aggregated CA;MRI

� �iron concentrations were determined from Eqs. [(3–5)].Linear calibrations between R2, R2* and iron wereestimated using univariate regression.

The background signal must be considered in thefitting procedure, and several techniques have beenused by other investigators (14,15,24,27). In additionto assuming a noise bias as described above, we com-pared R2 fitting methods proposed by other research-ers with our fitting strategy to evaluate the robustnessof our approach in the presence of background sig-nals. In general, the other techniques assume either a

MR Measurement of Hepatic Storage Iron 309

slow decaying component (14,24) or a constant(15,27) added to the signal decay, representing aweak signal contributed by other sources such as bileand blood. Based on the MSE data with the interechotime of 4 ms, we estimated R2 values with (i) the biex-ponential method, which uses summation of R2 fromthe fast and slow decay components weighted by theirpopulation densities, and (ii) the monoexponentialmethod, which contains the second component as aconstant (27).

Phantoms

To validate our methods, we conducted a phantomstudy similar to that of Ref. (30), but with differentpulse sequences. Phantoms were prepared, as in Jen-sen et al (30), with mixtures of various concentrationsof MnCl2 (0 to 0.675 mM) and iron oxide microspheres(0 to 0.1 mg Fe/mL), to mimic, respectively, the relax-ation effects of ferritin and hemosiderin iron found iniron-loaded liver. The iron oxide microspheres werecomposed of magnetite nanoparticles (UM3CN/5737,Bangs Laboratories, Fishers, Indiana), with a meanradius of 2.9 mm and an average of 4.4 � 10�9 mgFe/particle. A total of 36 cylindrical sample bottles(1.5 cm in diameter and 15 cm3 in volume) were pre-pared with concentrations

CD¼ m � 0:0074 mg Mn=g ; m ¼ 0;1; � � �5;

CA ¼ n � 0:02 mg Fe=g ; n ¼ 0;1; � � �5: (7)

suspended in 2% agarose gel and immersed in a �50mM MnCl2 aqueous solution bath. The MnCl2 wasadded to the bath both to suppress the bath signal(which could contaminate the sample signal due toartifacts) and to reduce magnetic fieldinhomogeneities.

The phantom images were acquired using the sameMSE sequence designed for the liver study. Six bottleswere imaged simultaneously during each scan.Because the molecular weight of Mn (54.9) is close toFe (55.8), the “total iron” concentration (CT) was sim-ply the sum of the Mn (CD) and the microsphere iron(CA) concentrations.

Finally, stimulated echoes may occur for MSEsequences with inaccurate RF pulse flip angles andimperfect slice selection profile (31). To minimizethese, an optimized MSE sequence was implementedon the same MR system with an increased slice selec-tion thickness for the refocusing pulses (19,30,32,33).While the in vivo iron measurement was based on thevendor’s default (nonoptimized) MSE sequence, weevaluated our method on phantoms with both nonop-timized and optimized MSE sequence to assess poten-tial systematic effects.

RESULTS

MRI Iron Measurement for Aqueous Phantoms

Figure 1 demonstrates the three sets of MSE signaldecay in phantoms that contain Mn only (0.015 mgMn/mL) and that contain both Mn and Fe (0.015 mg

Mn/mL, 0.06 mg/mL). As can be seen, the MSE sig-nal decay in phantoms with no microspheres (CA¼0)was independent of the interecho time. With micro-spheres present, the signal decay, in contrast,depended strongly on the interecho time. Solid linesare the global fits for the multiple decay curves.

The solvent of the microspheres, which is in propor-tion to the iron concentration, was found to influenceMRI signal decay, with a net R2 effect of 133.0 6 4s�1/(mg Fe/g) estimated as discussed in a previousstudy (30). To minimize the solvent’s influence on thephantom results, the phantom R2 and RR2 valueswere corrected by subtracting out this net R2 effect.

We noticed that in the present study, the estimatedrelaxivity of the MnCl2 phantom was approximately59.1 s�1/mM at 1.5T, which is lower than the pub-lished values of 74.2 s�1/mM measured with a singlespin echo sequence (22) and 71.4 s�1/mM measuredwith the MSE sequence (38). We believe this underes-timation to be mainly due to stimulated echoes intro-duced by the imperfect slice selection profile andreduced refocusing RF pulse flip angle (19,30–33).When the refocusing slice thickness is increased byan empirically determined factor of three, the relaxiv-ity then becomes consistent with the literature values.

The in vivo image data in the present study wereacquired using the vendor-supplied, nonoptimizedMSE sequence, i.e., without increasing the slice selec-tion thickness for the refocusing RF pulses. This dif-ference prompted investigation of the relationshipbetween RR2 and A values derived from the presentand optimized MSE sequences (32,33). A linearregression analysis (r¼0.99) verified that the stimu-lated echo effects contributed to an underestimationof RR2 (Fig. 2a) and an overestimation of A (Fig. 2b).The relationship between the measurements madewith the vendor-supplied and optimized MSE sequen-ces appears to be linear, suppoprting the feasibility ofcalibrating RR2 and A (as measured with the vendor-supplied sequence) for total and compositional ironquantifications. A least squares fit of Eq. [(3)] to thefull set of phantom results yields the calibrationparameters

Figure 1. Multiple MSE signal decay in phantoms (0.015 mgMn/mL) without and with Fe (0.06 mg/mL). Signal intensityis in natural log scale.

310 Tang et al.

a1 ¼ � 11:361:9ð Þ � 10�3 mg Mn =gð Þ;a2 ¼ 11:860:56ð Þ � 10�4s � mg Mn =gð Þ;a3 ¼ 0:7560:015Þms 3=2 � mg Fe =gð Þ:

(8)

Figure 3 demonstrates the estimation of Mn and Feconcentration maps for phantoms containing variousFe concentrations (0 to 0.1 mg/mL) and a fixed Mn

concentration (0.015 mg/mL) using the two-parame-ter method. The phantom in the center (red arrow)contains no Mn and Fe. The phantom on the top (yel-low arrow) contains Mn only. Figure 3b and 3c are Mnand Fe maps estimated by Eqs. [(4)] and [(5)], respec-tively. Figure 3d is the summation of Mn and Fe, sim-ulating the mapping of total iron concentration.

The MRI determined total iron concentrations esti-mated by Eqs. [(3–5)] and the calibration parameters(Eq. [(8)]) are strongly correlated (Pearson correlationcoefficient r¼0.99), and in excellent agreement withthe true iron concentrations (mean difference was 0.0mg Fe/g, and 95% limits of agreement were 0.006and �0.005 mg Fe/g).

In Vivo MRI Liver Iron Measurement

Figure 4 demonstrates the compositional iron concen-tration maps in a healthy (Fig. 4a) and an iron over-loaded liver (Fig. 4d). The liver ROI template isindicated by red arrow. The mean RR2 and A from thesix healthy subjects were 20.8 6 4.0 s�1 and0.0029 6 0.0020 ms3/2, respectively, representing thenormal range of relaxation values, with an assumedtotal storage iron concentration of 0.15 mg Fe/g wetweight.

A least squares fit of Eq. [(3)] to the total storageiron concentration determined by biopsy or biosus-ceptometry measurements (n¼20) yielded the calibra-tion parameters:

a1 ¼ � 0:2860:30ð Þ mg Fe =gð Þ;a2 ¼ 5:962:0ð Þ � 10�2s � mg Fe =gð Þ;

a3 ¼ 3363ð Þms �3=2 � mg Fe =gð Þ:(9)

The ferritin (Fig. 4b and 4e) and hemosiderin (Fig.4c and 4f) maps were calculated based on Eqs. [(4)]and [(5)] using the two-parameter (RR2, A) method.The total storage iron concentration map (sum of theferritin and hemosiderin) is shown in Figure 4g.

The total iron concentration calculated based on theR2 and R2* are demonstrated in Figure 4h and 4i,respectively, for the iron overload liver. The linear cali-brations are:

Figure 2. Comparison of RR2 and A in phantoms using thenonoptimized and optimized MSE sequences. a: Estimationof RR2 in phantoms with various Mn concentrations (0 to0.037 mg/mL). b: Estimation of A in phantoms with variousFe concentrations (0 to 0.1 mg/mL) and a fixed Mn concen-tration (0.015 mg/mL).

Figure 3. Estimation of Mn and Fe concentration maps from phantoms containing various Fe concentrations (0 to 0.1 mg/mL) and a fixed Mn concentration (0.015 mg/mL). a: Phantom image. b: Mn map. c: Fe map d: CT (FeþMn) map. The centerphantom (red arrow) contains no Mn and Fe. The top phantom (yellow arrow) contains Mn only. The color bar represents arange of 0 to 0.12 mg Fe/mL.

MR Measurement of Hepatic Storage Iron 311

CT R2 ¼ 0:13s � R2 � 2:5ð Þ mg Fe =gð Þ;CT R�2 ¼ 0:0135 s � R�2 � 0:5

� �mg Fe =gð Þ: (10)

There were only traces of hemosiderin iron in nor-mal liver as can be seen from the fact that varying theinterecho time had little to no effect on the signaldecay (Fig. 5). In the iron-loaded liver tissue, elevatedferritin and hemosiderin iron are evident. The signaldecay depends strongly on the interecho time (Fig. 5).The total iron in this specfic iron overload liver esti-mated by R2* (5.1 mg Fe/g) is greater than that fromthe two-parameter (RR2, A) method, from R2, and

from biopsy, or 2.6, 1.9, and 2.4 mg Fe/g wet weight,respectively.

Total storage iron, ferritin- and hemosiderin-likeiron concentrations in liver ROI were calculated for allsubjects. The mean total storage iron estimated by(RR2, A) and biopsy/biosusceptometry are 3.26 6 2.25and 3.34 6 2.32 mg Fe/g, respectively, for the 20patients. The mean total iron estimated by (RR2, A) fornormal subjects is 1.04 6 0.22 mg Fe/g. As shown inFigure 6a, a strong correlation (r¼0.95) was foundbetween the total iron measured by (RR2, A) and thatdetermined by biopsy or biosusceptometry. Strongcorrelations were also found between the R2-based

Figure 4. Compositional iron concentration maps in a healthy (a–c) and an iron overload liver (d–i, 2.4 mg Fe/g). a: Normalliver. b: Ferritin map. c: Hemosiderin map. d: Iron overload liver. e: Ferritin map. f Hemosiderin map. g: Total iron estimatedby RR2, A. h: Total iron estimated by R2. i: Total iron estimated by R2*. The liver ROI templates are indicated by red arrows.The color bar represents a range of 0 to 10 mg Fe/g wet weight.

312 Tang et al.

iron measurements and the true iron concentrations(r¼0.88). In contrast, the R2*-based iron measure-ments were saturated for high iron overload cases (>5 mg Fe/g), due to significant signal drop out at theearliest echo, and showed weak correlation with thetrue iron concentrations (r¼0.36). However, if thesehigh overload cases are excluded, the correlationincreases (r¼0.80) (data not shown).

Figure 6b shows the fractional hemosiderin ironconcentration as a function of the biopsy and biosus-ceptometry derived total iron concentration, suggest-ing that the proportion of the total storage ironexisting as hemosiderin grows as the total iron con-centration increases, which is consistent with theresults of the previous study (30).

Comparison of R2 and RR2 Measurements inPhantoms

The aggregation index A is independent of the Mnconcentrations (30), and as shown in Figure 2b, pro-portional to the microsphere iron concentration. Thedependencies of R2 and RR2 on the microsphere ironconcentration can be indirectly represented by theirrelationships with A (Fig. 7a). There is no correlationbetween RR2 and A (r¼0.06).

There are two sources of relaxations contributing toR2, that from the MnCl2, a diffuse, soluble compo-nent, R2_Mn, and that from the iron oxide micro-spheres, an aggregate, insoluble component, R2_IO,and one may plausibly write

R2 ¼ R2 Mn þ R2 IO : (11)

The dependency of R2 on A (r¼0.63) was used toanalyze the two R2 components. By subtracting theRR2 from R2, a linear curve fitting

Y ¼ 0:23 ms 1=2 � A (12)

was found to best represent the dependency part ofR2 on A, and can be used to estimate the R2_IO

component in R2. The R2_Mn can be estimated by sim-ply subtracting the R2_IO from R2.

As shown in Figure 7b, the R2 values (circles) areloosely distributed due to the presence of micro-spheres that affect R2 but not RR2. High correlations(r¼0.99) and excellent agreement were found betweenthe R2_Mn and RR2 (dots), indicating that the RR2

relaxation rate is mainly associated with the MnCl2,thus equivalent to the soluble component of R2, theR2_Mn.

Comparison of R2 and RR2 Measurements in Liver

Similarly, for iron overload liver, there are two sourcesof relaxations contributing to R2, that from the ferri-tin, R2_F, and that from the hemosiderin, R2_H, andthus

R2 ¼ R2 F þ R2 H: (13)

The dependencies of R2 and RR2 on the hemosideriniron concentrations can be estimated by their rela-tionships with A (Fig. 7c). Little correlation was foundbetween RR2 and A (r¼0.17), suggesting that therelaxation rates derived from the two forms of signaldecay are independent for the iron overloaded liver.The dependent part of R2 on A (r¼0.82), which is

Figure 5. Multiple MSE signal decay in the healthy and ironoverload livers. The liver regions for acquiring the signaldecays are indicated in Figure 4a and 4d.

Figure 6. Comparison of total iron concentration betweenthe MRI estimations using the two-parameter (RR2, A)method and the biopsy/biosusceptometry measurements. a:

Total iron quantification using the two-parameter method. b:

Fractional hemosiderin iron concentration (CA/CT).

MR Measurement of Hepatic Storage Iron 313

associated with the hemosiderin iron and may be esti-mated from the linear fit R2_H¼0.23 ms1/2 � A, wassubtracted from the R2 to calculate the R2_F

component.The comparisons between R2 versus RR2 and R2_F

versus RR2 are demonstrated in Figure 7d. Asexpected, the R2 values (circles) are loosely correlatedwith RR2 values (r¼0.41). There are two outliers of R2

values indicated by an arrow, representing two highhepatic iron overload cases, in which the proportionsof storage iron in the hemosiderin form are approxi-mately 95%. Nevertheless, with the subtraction ofR2_H, excellent correlations were found between theR2_F and RR2 (dots) values (r¼0.99), confirming thatthe RR2 relaxation rate is mainly associated with theferritin iron, and is equivalent to the R2_F componentin R2.

Comparison of R2 Measurements Using DifferentFitting Procedures

Finally, strong correlations were found between thepresent R2 estimates assuming a noise level as a fit-ting parameter in Eq. [(6)] and bi-exponential R2

measurements (r¼0.99) (14), and between the presentand the mono-exponential R2 measurements

assuming a constant bias component (r¼0.99) (27),which confirmed the robustness of our fitting strategyin the presence of background signal and noise bias,and verified the improvement of the total iron mea-surement using the two parameters (RR2, A) derivedfrom the same fitting strategy.

DISCUSSION

The proposed MRI method allows for the separatequantification of diffuse (ferritin-like) and aggregated(hemosiderin-like) storage iron in liver. The presenttwo-parameter (RR2, A) model, the modified MSEsequences that contain three different interechotimes, and the parallel imaging and motion suppres-sion gating system provided high quality images yield-ing accurate estimates of total as well ascompositional (ferritin- and hemosiderin-like) ironconcentrations in the liver. The validity and feasibilityof the two-parameter method is supported by bothphantom and in vivo human data in the presentstudy. Consistent with the results in Jensen et al (30)and Zuyderhoudt et al (39), the proportion of hepaticiron stored as ferritin, as determined by the two-pa-rameter method, decreases as the total storage ironconcentrations increase. The resulting MRI total

Figure 7. Separation of the R2 components in phantom and in liver. a: Phantom: the dependencies of R2 and RR2 on A. b:

Phantom: comparisons between RR2 and R2, and between RR2 and R2_Mn. c: Liver: the dependencies of R2 and RR2 on A. d:

Liver: comparisons between RR2 and R2, and between RR2 and R2_F. The arrow points to two high hepatic iron overload casescontaining high proportions of hemosiderin iron.

314 Tang et al.

hepatic iron measurement was highly correlated withiron concentration estimated from the biopsy or bio-susceptometry methods. The ability to separately mea-sure ferritin and hemosiderin iron may be clinicallyuseful to evaluate patients with iron overload and tomonitor the results of treatment with phlebotomy orchelation with higher specificity and sensitivity.

Note that the calibration parameters for phantomsare substantially smaller in magnitude than that forthe livers, due to the fact that ferritin and hemosid-erin are less efficient than MnCl2 and iron oxidemicrospheres in causing MRI signal decay. Therefore,the range of the absolute concentrations used in thephantoms is smaller than that of the iron overloadedlivers, although the magnetic and relaxation effectsare similar.

The established MRI iron quantification methods(12–15,17–19,22–24,27,28) provide total iron esti-mates, but do not separately determine ferritin ironlevels in patients with iron overload, for which ferritinmay constitute only a fraction of the total iron concen-tration. In our phantom study, varying the proportionof Mn and microspheres in phantom mixtures of dif-ferent CT may yield the same estimates for total ironusing the single-parameter (R2 and R2*) approach.The variability in the relative fraction of iron in dis-persed ferritin and clustered hemosiderin mayexplain, at least in part, the curvilinear relationshipbetween R2 and liver iron concentration in St Pierre etal (24). Compared with the present two-parametermethod, the single-parameter methods were lessstrongly correlated with the total iron concentration.

It should be noted that the two R2 components, R2_F

and R2_H, can be determined empirically by removingthe dependencies of R2 on the parameter A using thetwo-parameter method. The RR2 correlated excellentlywith the R2_F, indicating that it is mainly associatedwith the dispersed iron. This is validated by the phan-tom study, which further confirmed the capability ofthe two-parameter method to separately measure thetwo forms of storage iron, ferritin and hemosiderin,using the RR2 and A parameters. The conventional R2

depends strongly on the interecho time when hemo-siderin iron is presented, and it alone is not able toquantifty the ferritin and hemosiderin iron separately.

Motion artifacts, noise bias, slowly varying compo-nents from blood and bile, and partial volume effectscan contribute to nonmonoexponential signal decayand are, therefore, potential confounding factors forour two-parameter method. ECG triggering and respi-ratory navigator gating can be used to effectively cor-rect for the cardiac and respiratory motion. Recently,a breathhold multi-echo fast spin-echo (FSE) strategy(40) was proposed for measurement of R2 in the liverand heart, which reduced imaging time to a breath-hold duration of approximately 20 s. When calculat-ing iron concentration in liver, the blood vessels canbe excluded based on their image intensity in theMSE images. Overall, an advantage of the presentMSE acquisition strategy with multiple interechotimes is that it helps mitigate the confounding effectscaused by the background noise and slowly decayingcomponents, due to the specific interecho time

dependency predicted by Eq. [(1)]. The backgroundnoise defined in the two-parameter model alsoreduces the estimation error. Moreover, the shiftedMSE acquisition allows the first echoes to be acquiredat the same time, so that the signal decay from differ-ent acquisitions can be sampled at shorter times andscaled consistently to correct for the longitudinalmagnetization recovery variations or intra-scan differ-ences, minimizing the estimation error for the relaxa-tion parameters. As a result, the total iron quantifiedby the present two-parameter model showed the bestcorrelation with the true iron concentration in bothphantom and human studies in vivo.

In our study, the global fitting to estimate the fourparameters (S0, RR2, A, and r), and the monoexpo-nential fitting to estimate the three parameters (S0,R2/R2*, and r) are robust and rapid. There is no strictrequirement for initial parameter estimates. The fit-ting and estimates can be applied to signal decay inROI within an iron-loaded tissue, or on a pixel-by-pixel basis to map the relaxation parameters. Imageacquisition with higher spatial resolution can beimplemented for measuring iron concentration insmall tissues or organs.

The stimulated echo effects introduced by theimperfect slice selection profile and inaccurate RFpulse flip angle, based on the vendor’s default MSEimplementation, resulted in an underestimation ofRR2, and an overestimation of A. Regardless of thesources of the stimulated echoes, their effects on theestimation of RR2 and A can be calibrated to permitaccurate quantification of total as well as composi-tional (ferritin- and hemosiderin-like) iron concentra-tions. This feature demonstrates the feasibility ofapplying the two-parameter method in clinical set-tings having either the standard (default) or optimizedMSE sequences. However, the variations in calibrationparameters should be expected due to the variationsof RR2 and A estimated using different scanners andsequences (30). Nonetheless, although the two-param-eter method appears to be robust with respect to thestimulated echo effects, the refocusing slice thicknessand RF pulse flip angle should be consistent through-out a single study, and MSE sequences with minimalstimulated echo effects should be implemented acrossscanners for accurate R2-based iron quantification.

In conclusion, we report MRI measurement of dis-persed (ferritin-like) and aggregated (hemosiderin-like)hepatic storage iron in patients with transfusionaliron overload. This method may provide importantnew information about the partition of storage ironbetween ferritin and hemosiderin in other forms ofiron overload, and could potentially improve the accu-racy of conventional single-parameter (R2 or R2*) tech-niques for hepatic iron quantification in vivo. MRIdetection of increases in ferritin iron could potentiallyprovide a noninvasive early indicator of expansion ofthe cytosolic iron pool leading to an elevated risk ofiron-induced tissue toxicity in the liver, heart, andother tissues. Finally, MRI measurement of ferritiniron may offer a means to rapidly monitor the effectsof treatment of iron overload by phlebotomy or admin-istration of iron-chelating agents.

MR Measurement of Hepatic Storage Iron 315

ACKNOWLEDGMENTS

This work was perfomred at Columbia University.There is no conflict of interest with regard to thepublication.

REFERENCES

1. Brittenham GM, Badman DG. Noninvasive measurement of iron:report of an NIDDK workshop. Blood 2003;101:15–19.

2. Arosio P, Ingrassia R, Cavadini P. Ferritins: a family of moleculesfor iron storage, antioxidation and more. Biochim Biophys Acta2009;1790:589–599.

3. Chasteen ND, Harrison PM. Mineralization in ferritin: an efficientmeans of iron storage. J Struct Biol 1999;126:182–194.

4. De Domenico I, Vaughn MB, Li L, et al. Ferroportin-mediated mo-bilization of ferritin iron precedes ferritin degradation by the pro-teasome. EMBO J 2006;25:5396–5404.

5. Iancu TC. Ferritin and hemosiderin in pathological tissues. Elec-tron Microsc Rev 1992;5:209–229.

6. Miyazaki E, Kato J, Kobune M, et al. Denatured H-ferritin subu-nit is a major constituent of haemosiderin in the liver of patientswith iron overload. Gut 2002;50:413–419.

7. Wixom RL, Prutkin L, Munro HN. Hemosiderin: nature, formation,and significance. Int Rev Exp Pathol 1980;22:193–225.

8. Brittenham GM. Iron-chelating therapy for transfusional ironoverload. N Engl J Med 2011;364:146–156.

9. De Domenico I, Vaughn MB, Paradkar PN, Lo E, Ward DM,Kaplan J. Decoupling ferritin synthesis from free cytosolic ironresults in ferritin secretion. Cell Metab 2011;13:57–67.

10. Fleming RE, Ponka P. Iron overload in human disease. N Engl JMed 2012;366:348–359.

11. De Domenico I, Ward DM, Kaplan J. Specific iron chelators deter-mine the route of ferritin degradation. Blood 2009;114:4546–4551.

12. Bonkovsky HL, Rubin RB, Cable EE, Davidoff A, Rijcken TH,Stark DD. Hepatic iron concentration: noninvasive estimation bymeans of MR imaging techniques. Radiology 1999;212:227–234.

13. Clark PR, Chua-Anusorn W, St Pierre TG. Proton transverserelaxation rate (R2) images of liver tissue; mapping local tissueiron concentrations with MRI [corrected]. Magn Reson Med2003;49:572–575.

14. Clark PR, Chua-anusorn W, St Pierre TG. Bi-exponential protontransverse relaxation rate (R2) image analysis using RF field in-tensity-weighted spin density projection: potential for R2measurement of iron-loaded liver. Magn Reson Imaging 2003;21:519–530.

15. Clark PR, St Pierre TG. Quantitative mapping of transverse relax-ivity (1/T(2)) in hepatic iron overload: a single spin-echo imagingmethodology. Magn Reson Imaging 2000;18:431–438.

16. Ernst O, Sergent G, Bonvarlet P, Canva-Delcambre V, Paris JC,L’Hermine C. Hepatic iron overload: diagnosis and quantificationwith MR imaging. AJR Am J Roentgenol 1997;168:1205–1208.

17. Gandon Y, Olivie D, Guyader D, et al. Non-invasive assessment ofhepatic iron stores by MRI. Lancet 2004;363:357–362.

18. Jensen PD. Evaluation of iron overload. Br J Haematol2004;124:697–711.

19. Pell GS, Briellmann RS, Waites AB, Abbott DF, Lewis DP, JacksonGD. Optimized clinical T2 relaxometry with a standard CPMGsequence. J Magn Reson Imaging 2006;23:248–252.

20. Pierre TS, Chan P, Bauchspeiss K, et al. Synthesis, structure andmagnetic properties of ferritin cores with varying compositionand degrees of structural order: models for iron oxide deposits iniron-overload diseases. Coord Chem Rev 1996;151:125–143.

21. Poon CS, Henkelman RM. Practical T2 quantitation for clinical

applications. J Magn Reson Imaging 1992;2:541–553.22. St Pierre TG, Clark PR, Chua-Anusorn W. Single spin-echo pro-

ton transverse relaxometry of iron-loaded liver. NMR Biomed

2004;17:446–458.23. St Pierre TG, Clark PR, Chua-Anusorn W. Measurement and

mapping of liver iron concentrations using magnetic resonance

imaging. Ann N Y Acad Sci 2005;1054:379–385.24. St Pierre TG, Clark PR, Chua-anusorn W, et al. Noninvasive mea-

surement and imaging of liver iron concentrations using proton

magnetic resonance. Blood 2005;105:855–861.25. Storey P, Thompson AA, Carqueville CL, Wood JC, de Freitas RA,

Rigsby CK. R2* imaging of transfusional iron burden at 3T and

comparison with 1.5T. J Magn Reson Imaging 2007;25:540–547.26. Thomsen C, Wiggers P, Ring-Larsen H, et al. Identification of

patients with hereditary haemochromatosis by magnetic reso-

nance imaging and spectroscopic relaxation time measurements.

Magn Reson Imaging 1992;10:867–879.27. Wood JC, Enriquez C, Ghugre N, et al. MRI R2 and R2* mapping

accurately estimates hepatic iron concentration in transfusion-

dependent thalassemia and sickle cell disease patients. Blood

2005;106:1460–1465.28. Wood JC, Ghugre N. Magnetic resonance imaging assessment of

excess iron in thalassemia, sickle cell disease and other iron

overload diseases. Hemoglobin 2008;32:85–96.29. Jensen JH, Chandra R. Theory of nonexponential NMR signal

decay in liver with iron overload or superparamagnetic iron oxide

particles. Magn Reson Med 2002;47:1131–1138.30. Jensen JH, Tang H, Tosti CL, et al. Separate MRI quantification

of dispersed (ferritin-like) and aggregated (hemosiderin-like) stor-

age iron. Magn Reson Med 2010;63:1201–1209.31. Williams CF, Redpath TW, Smith FW. The influence of stimulated

echoes on contrast in fast spin-echo imaging. Magn Reson Imag-

ing 1996;14:419–428.32. Sammet CL, Swaminathan SV, Tang H, et al. Measurement and

correction of stimulated echo contamination in T(2)-based ironquantification. Magn Reson Imaging 2012 [Epub ahead of print].

33. Wu EX, Kim D, Tosti CL, et al. Magnetic resonance assessment ofiron overload by separate measurement of tissue ferritin and he-mosiderin iron. Ann N Y Acad Sci 2010;1202:115–122.

34. Vymazal J, Zak O, Bulte JW, Aisen P, Brooks RA. T1 and T2 offerritin solutions: effect of loading factor. Magn Reson Med1996;36:61–65.

35. Overmoyer B, McLaren C, Brittenham G. Uniformity of liver den-sity and nonheme (storage) iron distribution. Arch Pathol LabMed 1987;111:549–554.

36. Brittenham GM, Sheth S, Allen CJ, Farrell DE. Noninvasivemethods for quantitative assessment of transfusional ironoverload in sickle cell disease. Semin Hematol 2001;38(Suppl1):37–56.

37. Henkelman RM. Measurement of signal intensities in the pres-ence of noise in MR images. Med Phys 1985;12:232–233.

38. Ulmer JL, Mathews VP, Hamilton CA, Elster AD, Moran PR. Mag-netization transfer or spin-lock? An investigation of off-resonancesaturation pulse imaging with varying frequency offsets. AJNRAm J Neuroradiol 1996;17:805–819.

39. Zuyderhoudt FM, Sindram JW, Marx JJ, Jorning GG, van Gool J.The amount of ferritin and hemosiderin in the livers of patientswith iron-loading diseases. Hepatology 1983;3:232–235.

40. Kim D, Jensen JH, Wu EX, Sheth SS, Brittenham GM. Breath-hold multiecho fast spin-echo pulse sequence for accurate R2measurement in the heart and liver. Magn Reson Med2009;62:300–306.

316 Tang et al.