State of the Art Quantification of Liver Iron with MRI - Vendor Implementation and Available Tools

Abstract

The role of MRI to estimate liver iron concentration (LIC) for identifying patients with iron overload and guiding the titration of chelation therapy is increasingly established for routine clinical practice. However, the existence of multiple MRI-based LIC quantification techniques limits standardization and widespread clinical adoption. In this article, we review the existing and widely accepted MRI-based LIC estimation methods at 1.5T and 3T: signal intensity ratio (SIR) and relaxometry (R2 and R2*) and discuss the basic principles, acquisition and analysis protocols and MRI-LIC calibrations for each technique. Further, we provide an up-to-date information on MRI vendor implementations and available offline commercial and free software for each MRI-based LIC quantification approach. We also briefly review the emerging and advanced MRI techniques for LIC estimation and their current limitations for clinical use. Lastly, we discuss the implications of MRI-based LIC measurements on clinical use and decision-making in the management of patients with iron overload. Some of the key highlights from this review are as follows: 1) Both R2 and R2* can estimate accurate and reproducible LIC, when validated acquisition parameters and analysis protocols are applied, 2) Although the Ferriscan R2 method has been widely used, recent consensus and guidelines endorse R2*-MRI as the most accurate and reproducible method for LIC estimation, 3) Ongoing efforts aim to establish R2*-MRI as the standard approach for quantifying LIC, and 4) Emerging R2*-MRI techniques employ radial sampling strategies and offer improved motion compensation and broader dynamic range for LIC estimation.

CLINICAL NEEDS

Iron overload is a serious condition that usually arises either from increased gastrointestinal absorption of dietary iron (e.g., hereditary hemochromatosis) or from chronic blood transfusions (e.g., for sickle cell disease, β-thalassemia, or myelosuppression during chemotherapy).^1,2 As the body has no physiologic mechanism for clearing excess iron, extra iron accumulates in organs, primarily the liver, and ultimately leads to organ damage.³ Iron overload can result in significant morbidity and mortality if it is not effectively monitored and treated.^4,5 Liver iron concentration (LIC) is a reliable marker for total body iron accumulation and for guiding iron removal treatment.⁶ Although liver biopsy is historically the conventional reference standard for LIC estimation,⁷ this procedure is invasive and painful, has sampling variability,⁸ is associated with risks such as bleeding and infection and cannot be practically used for longitudinal LIC monitoring.⁹ Over the last two decades, magnetic resonance imaging (MRI) techniques have emerged to noninvasively measure LIC and closely monitor iron chelation therapy.¹⁰ However, several factors limit widespread clinical implementation of MRI-based methods, including: lack of standardization of the techniques, lack of proper clinical guidance and awareness of available resources for MRI-based LIC estimation, and lack of inline LIC estimation pipelines on MRI scanners. In children, additional challenges with MRI-based methods include long acquisition times and the need for breath holds.

This review aims to provide up-to-date information on widely accepted MRI techniques and available MRI vendor implementations and offline software and processing tools for LIC estimation. We also review emerging and advanced MRI techniques for LIC assessment and finally discuss the impact of MRI-based measurements on clinical use and decision making in the management of patients with iron overload.

EXISTING AND WIDELY AVAILABLE TECHNIQUES

The three widely accepted MRI-based techniques for clinical monitoring of LIC are signal intensity ratio (SIR) and relaxometry (R2/R2*). Paramagnetic nature of iron disrupts the homogeneity of the magnetic field in tissues and causing the water protons to dephase faster.^11,12 Thus, the presence of iron accelerates R2 and R2* relaxation rates, thereby causing signal loss on T2-weighted spin-echo and T2*-weighted gradient echo (GRE) images. Although MRI does not measure iron directly, it detects the effect of iron on water protons in the tissue of interest and quantifies LIC (mg/g or μmol/g dry tissue) indirectly using MRI-biopsy calibration methods.¹³

Signal Intensity Ratio (SIR)

In SIR methods, the signal intensity of the liver on GRE sequences is divided by the signal intensity of a reference tissue that does not accumulate iron such as skeletal muscle or fat.^14,15 The first and most widely recognized SIR method was that developed by Gandon et al. at 1.5 T.¹⁵ In this method, five breath-hold GRE sequences are obtained with constant repetition time (TR) – four with different echo times (TEs) and one with different flip angle to be able to quantify all levels of iron overload. SIR was calculated for each sequence and SIR-LIC regression equations were derived for various sequence combinations that provided accurate LIC estimation. More recent SIR studies from the same group acquired either five single-echo or one multi-echo GRE sequence with different TEs at 3T and derived SIR-LIC regression equations for each echo (not published).^16,17 LIC was then estimated using the regression equation that yielded the highest SIR-LIC correlation. Although the regression equations for estimating LIC from SIR data were not published by Gandon et al. at 1.5T and 3T, a different group, Alustiza et al. derived and published the below SIR-LIC calibration equation at 1.5 T using the same set of sequences as Gandon et al.

$$LIC(\mu mol/g)=e^{[5.808-\left(0.877\times S_{T2}\right)-\left(1.518\times S_{IW}\right)]}$$

where S_T2 and S_IW are the signal intensity measurements on T2 weighted and intermediate weighted GRE images, respectively.¹⁸

The only available software for estimating LIC using the SIR method is MRQuantif (imagemed.univ-rennes1.fr/en) that was developed by Gandon et al. at the University of Rennes, France. The software is freely accessible online or as a desktop application for performing image analysis and LIC estimation, but it lacks regulatory approval globally. The recommended acquisition is multi-echo GRE with 8–10 echoes with only the body coil selected; the surface coils should not be used for SIR method. The software allows the user to place circular regions-of-interest (ROIs) in liver, muscle and background (noise), and provides two SIR models (Gandon, Alustiza) to estimate LIC at 1.5T and one SIR model (d’Assignies) to estimate LIC at 3T and finally generates a LIC report. Figure 1 shows a representative example of estimating LIC by SIR method using MRQuantif software. This software can also estimate T2*/R2*-based LIC, which is discussed below under R2* section. In addition, the software provides an option to select the optimal model. In this case, initially, T2*/R2* is used but if the measured T2*< 2TE_min, or if the correlation coefficient (R²) between the T2* curve and the measured signal < 0.8, it automatically switches to SIR method to estimate LIC.

Figure 1.

Liver iron concentration (LIC) estimation using the web version of MRQuantif software for a representative sample dataset with iron overload scanned using 1.5 T GE Optima scanner (TE₁: 1.116 ms; ΔTE: 1.248 ms; 12 echoes). a) DICOM image viewer with region-of-interest (ROI) selection for liver, muscle, spleen and noise, b) MRI signal decay curves for each ROI, and c) the reported R2*, R2*-LIC (iron R2*) and SIR-LIC (iron SIR). The estimated LIC using SIR and R2* were 6.37 mg/g and 7.1 mg/g dry tissue, respectively. Note: MRQuantif does not have regulatory approval.

Some notable limitations of this method are that the SIR measurements depend on the acquisition parameters and tend to overestimate LIC measurements compared to biopsy and R2* methods.^{16,17,19–22} Although studies have demonstrated the reproducibility of SIR method at 3T using two different scanners (Philips and Siemens Healthineers), comprehensive reproducibility studies across field strengths, vendors and sites are warranted for reliable clinical adoption.^16,17 Further, the algorithm developed by Gandon et al. is currently used as a black-box approach for SIR-based LIC estimation since the regression equations used for estimating different LIC ranges are not published. Another limitation is that the use of body coils for homogenous signal sensitivity limits signal-to-noise ratio (SNR) and makes the SIR method incompatible with phased-array coils and parallel imaging for rapid acquisition.^17,23 Lastly, hepatic steatosis and fat within muscle fascial planes can affect the signal intensity of the liver and muscle respectively, and may confound SIR and LIC measurements.^20,22

Relaxometry

In relaxometry, a series of spin-echo or GRE images are acquired with increasing TEs, and the signal decay constants, i.e., relaxation times: T2 or T2* respectively are calculated.²⁴ These relaxation times are inversely related to the iron concentration, i.e., the greater the iron concentration, the faster the signal decay and the lower the T2 or T2* times. Current practice is to report relaxation rates, R2 (=1000/T2) or R2* (= 1000/T2*) instead of relaxation times, as the iron concentration and R2 or R2* are linearly related, i.e., the higher the iron concentration and the higher the R2*. T2 and T2* are expressed in milliseconds (ms), while R2 and R2* are expressed in 1/second (s⁻¹). Both R2 and R2* methods are suitable for calculating LIC estimates in clinical practice when performed using validated acquisition and analysis protocols as described in the following sub-sections.

R2-MRI

R2 is measured using a series of single spin-echoes with increasing TEs, each acquired after a separate excitation. R2-LIC conversion is estimated using the published curvilinear relationship between R2 and biopsy LIC at 1.5T, which was first derived by St. Pierre et al. using repetition time, TR = 2500 ms²⁵ and was later updated for TR = 1000ms.²⁶

$$R2\left(s^{-1}\right)=6.88+26.06{\left[LIC\right]}^{0.701}-0.438{\left[LIC\right]}^{1.402}(\text{TR}=2500\text{ms})$$

$$LIC(mg/g)={\left\{29.75-{\left[900.7-2.283e^{\frac{-0.19043+1.016385\text{ln}{R}_2}{0.983615}}\right]}^{\frac{1}{2}}\right\}}^{1.4265}(\text{TR}=1000\text{ms})$$

MRI Vendor Implementation

Various commercial and free software/tools are available to calculate R2 maps and estimate R2-LIC values, which are listed in Table 1. There are three primary categories of software that ultimately yield R2-LIC: inline software from some MRI vendors that can produce only R2 maps on the scanner console, offline software that outputs a R2 map and LIC estimate, and offline software that can only produce R2 maps. Some primary MRI scanner manufacturers have product sequences accompanied with postprocessing algorithms to generate inline R2 maps: T2Map/CartiGram (GE HealthCare, Waukesha, WI), StarQuant (Philips, Best, The Netherlands), and MapIt (Siemens Healthineers, Erlangen, Germany). However, these products are primarily indicated for cardiac or musculoskeletal applications, and currently no MRI vendor yet offers inline R2 mapping and LIC estimation for liver iron overload applications.

Table 1.

Offline Software for R2-LIC Estimation.

Offline Software	Acquisition	Signal Model(s) & Calibration	Utility & Features	Output(s)	Regulatory Approval
FerriScan Resonance Health33 (https://ferriscan.com/ferriscan/)	Single spin-echo TR: 1000ms; TE1/ΔTE: 6–18ms/3ms; 5 echoes; flip angle: 90°; slice thickness: 5–6mm; Acquisition time: ~ 10 minutes	Bi-exponential model: $S\left(TE\right)=S_f\left(0\right)e^{-R{2}_{f}TE}+S_s\left(0\right)e^{-R{2}_{s}TE}$ St. Pierre Calibration	Commercial; image analysis and LIC reporting is performed at ISO certified core laboratory	R2, LIC	FDA, EMA, MHRA, ATG
ParametricMRI (https://www.parametricmri.com/) Children’s Hospital of Philadelphia, Philadelphia, PA, USA34	Single spin-echo Recommended: TR: 2500ms; TE1/ΔTE: 6–18ms/3ms; 5 echoes; flip angle: 90°; slice thickness: 5–6mm Acquisition time: ~ 20 minutes	Monoexponential plus constant offset: $S\left(TE\right)=S_0{e}^{-TE/T2}+C$ Linear monoexponential: $S\left(TE\right)=S_0{e}^{-TE/T2}$ Biexponential: $SE\left(TE\right)=S_{0\left(1\right)}{e}^{-TE/T{2}_{\left(1\right)}}+S_{0\left(2\right)}{e}^{-TE/T{2}_{\left(2\right)}}+C$ St. Pierre Calibration	Free standalone software; multiple signal models and R2-LIC calibrations; semi-automatic liver segmentation; batch processing.	R2, R2*, LIC, T1
Qmass (https://medisimaging.com/software-solutions/medis-suite-mr/) Medis Medical Imaging, Netherland35,36	Acquisition parameters not reported	Monoexponential $S\left(TE\right)=S_0{e}^{-TE/T2}$	Commercial standalone software; automatic contour detection or manual ROI; marketed for both heart and liver	R2, T2*, T1	FDA, EMA, ANVISA, TGA, PMDA, MFDS
Mrmap (https://sourceforge.net/projects/mrmap/ Cardiac MRI Unit, Franz-Volhard-Klinik, Charité University Medicine, Berlin, Germany32	Acquisition parameters not reported	Monoexponential $S\left(TE\right)=S_0{e}^{-TE/T2}$	Free standalone software; platform independent	R2, T2*, T1

Abbreviations: S, measured signal; TE, echo time; S_f, signal amplitude of fast relaxation component; R2_f, transverse relaxation rate corresponding to S_f; S_s, signal amplitude of slow relaxation component; R2_s, transverse relaxation rate corresponding to S_s; S₀, initial signal intensity; C, offset to account for noise; LIC, liver iron concentration

Region(s), regulatory body: United States of America, Food and Drug Administration (FDA); European Union, European Medicine Agency (EMA); Australia, Therapeutic Goods Administration (TGA); United Kingdom, Medicines and Healthcare products Regulatory Agency (MHRA); Brazil, Brazilian Health Regulatory Agency (ANVISA); Japan, Pharmaceuticals and Medical Devices Agency (PMDA); South Korea, Ministry of Food and Drug Safety (MFDS)

Offline Software

R2-LIC

To-date, the St. Pierre method is the only regulatory-approved R2-based LIC reporting technique, and is commercially available as FerriScan^®, marketed by Resonance Health (http://www.resonancehealth.com/). FerriScan requires acquisition of five T2-weighted single spin-echo sequences while free-breathing with constant TR and increasing TEs (Table 1) with placement of a saline bag within the field-of-view to correct for instrumental gain drift between sequences acquired at different TEs.^25,27 The acquired images are securely sent to Resonance Health for post-processing and image analysis that includes gain drift and respiratory motion correction, background noise subtraction, and bi-exponential modeling to generate a quantitative liver R2 map and FerriScan LIC report.

FerriScan acquisition can be performed on most of the MRI vendors (Siemens, Philips, GE, and others) at 1.5T and recently it has also been marketed for 3T scanners. The image analysis and generation of a FerriScan LIC report can take up to 2 business days. Recently, Resonance Health has launched another FDA-cleared technique, FerriSmart, which uses artificial intelligence for image analysis and generates a LIC report in real time.²⁸ A recent study has shown FerriSmart LIC estimates to be in good agreement with FerriScan, hence demonstrating the potential for providing clinicians with instant LIC results for making diagnostic and treatment decisions for patients.²⁸ Figure 2 shows representative examples of FerriScan and FerriSmart estimates in patients with iron overload. Both FerriScan and FerriSmart report the average LIC of a patient, while FerriScan provides the liver R2 map along with its histogram distribution within the selected ROI in liver. FerriSmart neither displays a liver R2 map nor an average R2 value.

Figure 2.

Example LIC reports from a) FerriScan and b) FerriSmart in a 13-yr old female and an 8-yr-old male diagnosed with sickle cell disease, respectively. FerriScan displays liver R2 map and histogram distribution of R2 values and reports the average LIC, mean ± standard deviation (SD) of R2 values, whereas FerriSmart only provides an average LIC.

Some limitations of the St. Pierre R2 methods are long acquisition time of ~10–15 minutes and ghosting artifacts due to free-breathing acquisitions, which can hinder accurate R2 and LIC quantification.^21,29 Further, FerriScan and FerriSmart R2 services are expensive and may or may not be reimbursable depending on the insurance company and country. Thirdly, as the R2-LIC relationship is curvilinear and plateaus at high LIC, the R2 methods can be insensitive for monitoring longitudinal changes in patients with severe iron overload.²² Lastly, the effect of concomitant presence of fat on R2 estimation is not well understood,³⁰ and to our knowledge, there is no study investigating the effects of steatosis on R2-LIC estimation. Nonetheless, despite these limitations, the FerriScan R2-LIC protocol and calibration are robust to intermachine variability, standardized,²⁷ and increasingly being used as a reference standard for validating other LIC estimation methods.^27,29,31

Alternative to FerriScan, parametric MRI (www.parametricmri.com) (pMRI) is a free LIC reporting tool that includes all post-processing and image analysis steps necessary to estimate LIC from spin-echo images acquired at different TEs. pMRI provides a selection for the user to choose the T2/R2 fitting model (all models available are listed in Table 1) for generating R2 maps, performs semi-automated liver segmentation, and estimates LIC using the St. Pierre R2-LIC calibration. Figure 3 shows a representative example of the results generated using pMRI for R2-LIC estimation.

Figure 3.

Representative LIC analysis of a 6-yr old female with mild iron overload using parametric MRI (pMRI). Dataset consisted of 5 spin-echo images (TE₁: 6 ms, ΔTE: 3ms, TR 1000 ms) acquired on a 1.5T Siemens MAGNETOM Avanto MRI scanner. a) The whole-liver ROI (red mask) overlayed on a magnitude image was selected semi-automatically by outlining the liver and then thresholding R2 values to exclude blood vessels and noisy bright pixels. b) Signal decay of spin echo acquisition obtained at five different echo times were fitted with a biexponential model, where the black-boxed legend displays fast (R2) and slow (R2₍₂₎) relaxivity values. c) The estimated fast component R2 map (sec⁻¹) is shown in grayscale with a horizontal color bar. d) Report table containing average LIC, T2 and R2 values, correlation coefficient (R²) of fit and ROI volume is displayed. Note: pMRI does not have regulatory approval. In this example, a biexponential R2 fitting is shown but pMRI can also perform monoexponential R2 fitting and generate corresponding results. The legend box in the signal fit was enlarged, and the R2 image was cropped and its color bar was adjusted for better visualization.

R2 Parametric Maps

R2 maps can be generated offline using commercial or free software (Qmass, Mrmap) listed in Table 1, however these software are not marketed for liver but for cardiac R2 estimation.³² Nevertheless, LIC can be estimated by drawing liver ROIs on the R2 maps and calculating mean R2 value, and manually computing LIC from the mean R2 value via the St. Pierre R2-LIC calibration equation.

R2*-MRI

The most widely used clinical sequence for quantifying R2* is a multi-echo GRE performed in a single breath-hold. The TEs of the GRE sequence should be selected in such a way that the first echo (TE₁) and the echo spacing (ΔTE) should be as short as possible (~1 ms) and the number of echoes should be 6–12 to capture sufficient signal decay and reliably measure normal to high LIC up to 25 mg/g.^22,37

Conventionally, 2D multi-echo GRE sequences were used to calculate liver R2* and to derive R2*-biopsy LIC calibrations.^31,37 Recently, 3D multi-echo GRE has become a more common acquisition for R2* estimation because of the availability of vendor sequences that can provide whole liver coverage in a breath-hold while producing inline R2* maps on the scanner console. Further, recent studies demonstrated a strong correlation between the R2* values obtained with 3D multi-echo GRE with those obtained with 2D GRE, hence making 3D GRE methods interchangeable with 2D methods for R2*-based LIC estimation.^38,39

To estimate LIC, the obtained GRE images undergo a series of postprocessing steps. This includes fitting a signal model to estimate R2*, selecting ROI(s), and finally R2*-LIC conversion using published R2*-biopsy LIC calibration equations. Alternatively, estimated R2* maps can be converted to LIC maps, and then ROI selection can be performed for reporting mean LIC. In the last two decades, multiple calibration curves have been published to estimate LIC from R2* and are summarized in Table 2. Although these studies used slightly different acquisition parameters and monoexponential fitting models (Table 2), the published R2*-biopsy LIC calibrations were not statistically different as long as the acquisition parameters were comparable.^{16,29,31,37,40–42} Some initial R2*-LIC calibrations derived using longer first echo times (2.2 ms and 4 ms) were significantly different and are not included in Table 2.^43–45

Table 2.

Published R2*-LIC calibration curves in literature at different field strengths along with the acquisition parameters and signal models.

Calibration Method	Acquisition	Signal Model	LIC Estimation (mg /g dry tissue)
1.5T
Wood42 (R2* vs biopsy-LIC)	2D Multi-echo GRE TR: 25ms; TE1/ΔTE: 0.8ms /0.25ms; 16 echoes; flip angle: 20°; slice thickness: 15 mm;	Monoexponential plus constant offset; magnitude fitting $S\left(TE\right)=S_0{e}^{-TE.{R2}^{*}}+C$	$LIC=0.0254\left(R{2}^{*}\right)+0.202$
Hankins37,66 (R2* vs biopsy-LIC)	2D Multi-echo GRE TR: 200ms; TE1/ΔTE: 1.1ms/0.8ms; 20 echoes; flip angle: 25°; slice thickness: 10mm;	Monoexponential with noise subtraction and truncation; magnitude fitting $S_{\mathit{\text{corr}}}\left(TE\right)=\sqrt{{\left(S\left(TE\right)\right)}^2-N^2}$ $S_{\mathit{\text{corr}}}\left(TE\right)=S_0{e}^{-TE.{R2}^{*}}$	$LIC=0.028\left(R{2}^{*}\right)-0.454$
Garbowski31,67 (R2* vs biopsy-LIC)	2D Multi-echo GRE TE1/ΔTE: 0.93ms/0.8ms; 20 echoes; slice thickness: 10mm	Monoexponential plus constant offset with truncation; magnitude fitting $S\left(TE\right)=S_0{e}^{-TE.{R2}^{*}}+C$	$LIC=0.032\left(R{2}^{*}\right)-0.14$
Henninger40 (R2* vs biopsy-LIC)	2D Multi-echo GRE TR: 200ms; TE1/ΔTE: 0.99ms/1.41ms; 12 echoes; flip angle: 20°; fat suppression; slice thickness: 10mm	Monoexponential plus constant offset with truncation, magnitude fitting $S\left(TE\right)=S_0{e}^{-TE.{R2}^{*}}+C$	$LIC=0.024\left(R{2}^{*}\right)+0.277$
Jhaveri41 (R2* vs FerriScan-LIC)	3D Multi-echo GRE TR: 12ms; TE1/ ΔTE: 1ms/1.4ms; flip angle: 6°; slice thickness: 4mm;	Joint fat-R2* model, complex/magnitude multi-step fitting $S\left({\rho }_W,{\rho }_F,{\varphi }_o,f_B,R_2^{*}\right)=\left({\rho }_W+{\rho }_F\sum _{p=1}^6{\alpha }_p{e}^{i2\pi f_{F,p}TE}\right)e^{i\left({\varphi }_0+2\pi f_{B}TE\right)}{e}^{-R_2^{*}TE}$	$LIC=0.0266\left(R{2}^{*}\right)$
Hernando29 (R2* vs FerriScan-LIC)	3D Multi-echo GRE TR: 9.6–12.6ms; TE1/ ΔTE: 0.7–0.8ms; 2 echo trains; 12 echoes; flip angle: 12°; slice thickness: 8mm;	Joint fat-R2* model, complex fitting; no fat correction for R2*> 500s−1 $S\left({\rho }_W,{\rho }_F,{\varphi }_o,f_B,R_2^{*}\right)=\left({\rho }_W+{\rho }_F\sum _{p=1}^6{\alpha }_p{e}^{i2\pi f_{F,p}TE}\right)e^{i\left({\varphi }_0+2\pi f_{B}TE\right)}{e}^{-R_2^{*}TE}$	$LIC=0.02603\left(R{2}^{*}\right)-0.16$
2.89T
Hernando29 (R2* vs FerriScan-LIC)	3D Multi-echo GRE TR: 6ms; TE1/ΔTE: 0.6ms; 2 echo trains; 8 echoes; flip angle: 9°; slice thickness: 8mm;	Joint fat-R2* model, complex fitting; no fat correction for R2*> 1000s−1 $S\left({\rho }_W,{\rho }_F,{\varphi }_o,f_B,R_2^{*}\right)=\left({\rho }_W+{\rho }_F\sum _{p=1}^6{\alpha }_p{e}^{i2\pi f_{F,p}TE}\right)e^{i\left({\varphi }_0+2\pi f_{B}TE\right)}{e}^{-R_2^{*}TE}$	$LIC=0.01400\left(R{2}^{*}\right)-0.03$
3.0T
D’Assignies16 (R2* vs biopsy LIC)	2D Multiecho GRE TR:12ms; TE1/ΔTE:1.15 or 1.23ms; 11 echoes; FA: 20°; slice thickness: 7mm	Monoexponential with noise subtraction, magnitude fitting $S_{\mathit{\text{corr}}}\left(TE\right)=\sqrt{{\left(S\left(TE\right)\right)}^2-N^2}$ $S_{\mathit{\text{corr}}}\left(TE\right)=S_0{e}^{-TE.{R2}^{*}}$	$LIC=0.0176\left(R{2}^{*}\right)+0.42$ *The actual relationship is reported in $\frac{\mu mol}{g}$
Hernando29 (R2* vs FerriScan-LIC)	3D Multiecho GRE TR: 6.0–7.0ms; TE1/ΔTE: 0.6–0.7ms; 2 echo trains; 8 echoes; flip angle: 9°; slice thickness: 8mm;	Joint fat-R2* model, complex fitting; no fat correction for R2*> 1000s−1 $S\left({\rho }_W,{\rho }_F,{\varphi }_o,f_B,R_2^{*}\right)=\left({\rho }_W+{\rho }_F\sum _{p=1}^6{\alpha }_p{e}^{i2\pi f_{F,p}TE}\right)e^{i\left({\varphi }_0+2\pi f_{B}TE\right)}{e}^{-R_2^{*}TE}$	$LIC=0.01349\left(R2*\right)-0.03$

Abbreviations: S, measured signal; TE, echo time; S₀, initial signal intensity; C, offset to account for noise; Scorr, corrected signal after noise subtraction; N, estimated noise; ρ_W, amplitude of water signal; ρ_F, amplitude of fat signal; Φ₀, initial phase; f_B, frequency shift due to local magnetic field inhomogeneities; f_F,p, known frequencies for the multiple fat peaks relative to the water peak; α_p, relative amplitudes of the multiple fat peaks such that $\sum _{p=1}^6{\alpha }_p=1$.^49,51

Hepatic steatosis is a confounding factor for R2* estimation as the presence of fat introduces oscillations, and the signal evolution cannot be accurately represented by a simple monoexponential model and requires a model containing both R2* and fat fraction.^46,47 Over the last decade, complex signal modeling techniques that incorporate fat-water separation and R2* mapping have emerged for simultaneous estimation of hepatic R2* and proton density fat fraction (PDFF). Recently, Hernando et al. performed a multi-center, multi-vendor reproducibility study based on a centralized offline R2* reconstruction incorporating a common joint fat-R2* signal model and reported R2*-LIC calibration curves for 1.5T, 2.89T and 3T using FerriScan LIC as a reference standard.²⁹ However, at high iron overload (R2*> 500 s⁻¹ at 1.5T and > 1000 s⁻¹ at 3T), the study discarded fat correction from the signal model to avoid instabilities in model fitting for reliable R2* estimation. The study reported that the calibrations were similar to previously published R2*-biopsy LIC calibrations and were not significantly different between children and adults.²⁹ In a follow-up study from the same group, agreement between vendor-provided inline R2* maps (GE HealthCare, Philips, Siemens Healthineers) and the centralized R2* reconstruction was assessed.⁴⁸ Although the vendor-provided R2* maps showed good agreement for mild iron overload, the study reported that at 1.5T, the upper limits of R2* ranges of agreement were approximately 500, 375, and 330 s−1 (i.e., LICs of 12.9, 9.6, and 8.4 mg/g) for GE, Philips, and Siemens reconstructions, respectively.⁴⁸ Similarly, at 3T, the upper limits of R2* agreement were approximately 700 and 800 s−1 (i.e., LICs of 9.4 and 10.8 mg/g) for GE and Philips, respectively. Within the R2* agreement range, vendor R2*-LIC calibrations were not significantly different from the centralized calibration.⁴⁸ However, none of the vendor reconstructions could accurately evaluate severe iron overload, likely in part due to different acquisition parameters used (longer TEs, higher in-plane resolution, monopolar echoes) compared to settings optimized for vendor R2* reconstructions.

R2*-MRI has major advantages over R2-based methods due to its ability to provide full liver coverage without motion artifacts using a single breath-hold scan. Further, in contrast to the curvilinear R2-LIC relationship, the R2*-LIC relationship is linear for the entire pathophysiological range of LIC, making R2* measurements desirable for clinical assessments.⁴² Moreover, a recent review and guidelines article from European Society of Gastrointestinal and Abdominal Radiology (ESGAR) and Society of Abdominal Radiology (SAR) groups concluded that R2* based LIC quantification is the most practical method for accurate and reproducible quantification of LIC.³⁰

MRI Vendor Implementation

Table 3 summarizes the vendor sequences and software/tools available for R2* and R2*-LIC estimation. Over the last decade, MRI vendors have developed software such as FFQ (Canon Medical Systems Corporation, Otawara, Japan), IronQuant/IDEAL-IQ (GE HealthCare, Waukesha, WI), mDIXON-Quant/StarQuant (Philips, Best, The Netherlands), and LiverLab (Siemens Healthineers, Erlangen, Germany), that acquire 2D/3D multi-echo GRE images and produce R2* maps on the scanner console. ROI measurements from these R2* maps can be converted to LIC units using a suitable calibration curve. However, to-date, none of the MRI vendor implementations are FDA approved yet for inline reporting of R2*-based LIC estimation on the scanner console.

Table 3.

MRI Vendor Implementation and Offline Software for R2*-LIC Estimation.

MRI Vendor/ Offline Software	Acquisition	Signal Model & Calibration	Outputs & Features	Regulatory approval
MRI Vendor Implementation
Fat Fraction Quantification (FFQ), Canon Medical Systems (https://global.medical.canon/products/magnetic-resonance/good-to-know#fat)	3D multi-echo GRE 1.5T/3T: TR: 7.8ms; TE1/ΔTE: 1.2ms/1.0ms; 6 echoes; flip angle: 3–5°; slice thickness: 6–8mm; bandwidth: 1302 Hz/pixel; acquisition time: 20s	Joint fat-R2* model49	R2*, PDFF, water, fat, in-phase, out-phase	FDA
IDEALIQ, GE HealthCare	3D multi-echo GRE 1.5T: TR: 9.6ms; TE1/ ΔTE: 0.7ms/0.7ms; 12 echoes; flip angle: 12°; slice thickness: 8mm; FOV: 40×32cm2; matrix: 144×128×32; voxel size: 2.8×3.1×8mm3; acceleration factor: 2; acquisition time: 21.2s 3T: TR: 6ms; TE1/ ΔTE: 0.6ms/0.6ms; flip angle: 9°; 8 echoes; slice thickness: 8mm; FOV: 40×32cm2; matrix: 144×128×32; voxel size: 2.8×3.1×8mm3; acceleration factor: 1.5; acquisition time: 17.3s	Joint fat-R2* model	R2*, PDFF, water, fat, in-phase, out-phase images, B0 field map	FDA, EMA, NMPA
IronQuant, GE HealthCare	Same as IDEALIQ, GE HealthCare	Joint fat-R2* model Hernando29 calibration	LIC, R2*, PDFF, water, fat, in-phase, out-phase images, B0 field map Inline LIC Map	Investigational Device
mDIXON-Quant Philips	3D multi-echo GRE 1.5T SENSE: TR: 5.3ms; TE1/ ΔTE: 0.92ms/0.7ms; 6 echoes; flip angle: 5°; slice thickness: 6mm; FOV: 400×350×231mm3; matrix: 132×118×38; reconstructed 192×192×77; voxel size: 3×3×6mm3; reconstruced 2.1×2.1×3mm3; acceleration factor: 2; acquisition Time: 12.1s 3T SENSE: TR: 5.6ms; TE1/ΔTE: 0.97ms/0.7ms; 6 echoes; flip angle: 3°; slice thickness: 6mm; FOV: 400×350×231mm; matrix: 160×140×38; reconstructed 192×192×77; voxel size: 2.5×2.5×6mm; reconstructed: 2.1×2.1×3.0mm; acceleration factor: 2; acquisition time: 15.6s	Joint fat-R2* model	R2*, T2*, PDFF, water, fat, in-phase, out-phase images, B0 and fit error maps. Compressed SENSE and SmartSpeed reconstruction * B0 and fit error maps are only available via research license. * Liver volume segmentation via the MR Liver Health package.	FDA, EMA, HSA
StarQuant, Philips	2D multi-echo GRE 1.5T: TR: 15ms, TE1/ ΔTE: 1.12ms/0.8ms; 16 echoes; flip angle: 25°; slice thickness: 5mm; FOV: 350×350mm; matrix: 140×140×10; voxel size: 2.5×2.5×5mm; acceleration factor: 1.5; acquisition time: 14.7s 3.0T: TR: 16ms; TE1/ΔTE: 1.19ms/0.9ms; 16 echoes; flip angle: 25°; slice thickness: 5mm; FOV: 350×350mm; matrix: 176×176×10; voxel size: 2×2×6mm; acceleration factor: 2; acquisition time: 15.1s *No default 3D acquisition protocols are provided	Monoexponential with noise floor correction	R2*, T2* Compatible with SENSE parallel imaging, Compressed SENSE and SmartSpeed reconstruction	FDA, EMA, HSA
LiverLab, Siemens Healthineers Multi-echo Dixon (“q-dixon”)	3D multi-echo GRE 1.5T: TR: 9ms; TE1/ ΔTE: 1.1ms/1.19ms; 6 echoes; flip angle: 4°; slice thickness: 3.5mm; matrix: 160×128; Bandwidth: 1100Hz/pixel; acceleration factor: 3; acquisition time: 15s 3T: TR: 9ms; TE1/ ΔTE: 1.1ms/1.23ms; 6 echoes; flip angle: 4°; slice thickness: 3.5mm; matrix: 160×128; Bandwidth: 1100Hz/pixel; acceleration factor: 4; acquisition time: 13s For higher iron levels, 1.5T & 3T: higher Bandwidth (1400Hz/pixel), minimized echo times (TE1/ ΔTE: 0.9ms/1ms), higher flip angle (6°), acceleration factor: 3; acquisition time: 15s.	Joint fat-R2* model	R2*, PDFF, water, fat, goodness of fit, original echoes User scan assistance, liver segmentation (from separate high-resolution 2pt Dixon scan if available), inline ROI, inline R2* & PDFF report	FDA
LiverLab, Siemens Healthineers Multi-echo Dixon works-in-progress (WIP) package	3D multi-echo GRE Same as for LiverLab product with modification for higher iron levels Breath-hold 3D Cartesian or free-breathing 3D stack-of-radial	Joint fat-R2* model Configurable R2*-LIC calibration, signal model adaptations for high R2*	Additional LIC map & report Radial self-gating	Research software
Offline software: R2*-LIC
LiverMultiscan (https://www.perspectum.com/our-products/livermultiscan/, Perspectum Diagnostics, Oxford, UK60,68	3D multi-echo GRE TR: 10ms; TE1/ΔTE: 1.1ms/1.0ms; 6 echoes; flip angle: 6°; slice thickness: 10mm; FOV: 35–40cm; matrix:128×128; acquisition time: 15s	Joint fat-R2* model, *calibration not reported	R2*, PDFF, LIC, iron corrected T1, automated whole liver segmentation Commercial; standalone software	FDA, EMA, HSA
MRQuantif (https://imagemed.univ-rennes1.fr/en), Prof. Yves Gandon, University of Rennes, Rennes, France20,61	2D multi-echo GRE (8–10 echoes) preferred Various GRE protocols: (https://imagemed.univ-rennes1.fr/en/mrquantif/protocols.php)	Eight different signal models to calculate T2* (https://imagemed.univ-rennes1.fr/en/mrquantif/quantif); Calibrations: Anderson, Woods, Hankins, Garbowski, Henninger	R2*, PDFF, LIC, SIR Non-commercial; standalone software; platform independent; online and desktop versions	No regulatory approval
Parametric MRI (pMRI) https://www.parametricmri.com/ Dmitry Khrichenko, Children’s Hospital of Philadelphia, Philadelphia, PA, USA69,70	2D Multi-echo GRE (minimum of 2 echoes) Wood et al. 2005 TR: 25ms; TE1/ΔTE: 0.8ms /0.25ms; 16 echoes; FA: 20°; slice thickness: 15 mm Henninger et al. 2015 2D Multi-echo GRE TR: 200ms; TE1/ΔTE: 0.99ms/1.41ms; 12 echoes; FA: 20°; fat suppression; slice thickness: 10mm	Monoexponential, monoexponential plus constant offset or bi-exponential; supports truncation; Calibrations: Anderson, Woods, Hankins, Garbowski, Henninger, Hernando (1.5T and 3.0T)	R2*, LIC, R2, T1, semi-automated whole liver segmentation Non-commercial; standalone software; batch processing	No regulatory approval
http://www.isodense.com/ic, Jose Michel Kalaf Research Institute, Campinas, Brazil71	Multi-echo GRE Acquisition parameters not reported	Monoexponential with truncation; Garbowski calibration	R2*, LIC Non-commercial; standalone software; online version	No regulatory approval
Offline software: R2* Parametric Maps
QLiver or QP-Liver (https://quibim.com/products/qp-liver/), QUIBIM, Valencia, Spain)20,62,63	Multi-echo GRE Acquisition parameters not reported	Joint fat-R2* model, complex fitting	R2*, PDFF, automated whole liver segmentation Commercial; standalone software;	Approval in Spain
MRmap (https://sourceforge.net/projects/mrmap/), Cardiac MRI Unit, Franz-Volhard-Klinik, Charité University Medicine, Berlin, Germany32	Multi-echo GRE Acquisition parameters not reported	Monoexponential; monoexponential plus constant offset	R2*, T2, T1 Non-commercial; standalone software; platform independent
QMass or QMAP (https://medisimaging.com/software-solutions/medis-suite-mr/), Medis Medical Imaging, Netherland35	Multi-echo GRE Acquisition parameters not reported	Monoexponential	R2*, T2, T1 Commercial; standalone software; automatic contour detection or manual ROI	FDA
CVI42 or CMR42 (https://www.circlecvi.com/cvi42), Circle Cardiovascular Imaging Inc., Calgary, Canada64,65	Multi-echo GRE Acquisition parameters not reported	Monoexponential; monoexponential plus constant offset; Monoexponential with truncation; Monoexponential with noise floor subtraction.	T2*, T2 Commercial; standalone software or plugin; can integrate with MRI vendor servers	FDA
CMRtools (Thalassemia Tools), Cardiovascular Imaging Solutions, London, UK64,65,72	Multi-echo GRE Acquisition parameters not reported	Monoexponential	T2* Commercial; standalone software or plugin;	FDA

Region(s), regulatory body: United States of America, Food and Drug Administration (FDA); European Union, European Medicine Agency (EMA); Singapore, Health Sciences Authority (HSA); China, National Medical Products Administration (NMPA)

Canon Medical

The Canon Medical Fat Fraction Quantification (FFQ) feature provides R2* maps that implicitly account for confounders such as fat. The feature is available on both 1.5T and 3T systems. Automated scan assistance is available to help with volume prescription. Data are acquired using a multi-echo, 3D GRE sequence. Bipolar readouts are used to reduce echo time spacing as compared to a monopolar (flyback) readout. Canon’s SPEEDER (SENSE-based) parallel imaging technique is compatible with FFQ to further reduce the scan time. Full 3D whole-liver coverage is achieved in a single breath-hold.

The acquired data are processed inline to generate the R2* map. The PDFF map, water image, fat image, in-phase (IP) image, and opposed-phase (OP) image may also be output. Processing is performed using a confounder-corrected, hybrid complex-magnitude fitting approach that offers robustness to B0 field heterogeneity¹²⁶ and reduces sensitivity to confounding effects while maintaining accurate R2* estimates and PDFF estimates over the full 0–100% range. An ROI may be drawn on the host computer or workstation to estimate the average R2* value (1/s) within the region. Figure 4 below shows an example R2* map and PDFF map reconstructed inline and displayed on the scanner host console.

Figure 4.

Automatic inline reconstruction of the (left) R2* map (s⁻¹) and (right) PDFF map (%) in a 42-yr old male using Canon Medical’s Fat Fraction Quantification (FFQ). ROI measurements (yellow circles) indicate an average R2* of 115 s⁻¹ and PDFF of 3.42%. Data are acquired on a 3T Vantage system (Canon Medical Systems Corporation, Otawara, Japan) using a 3D gradient-echo, six-echo sequence. Maps may be shown in greyscale or color (not shown).

GE HealthCare

GE HealthCare has developed an advanced technical software module (ATSM) for investigational purposes known as IronQuant which provides an on-scanner solution for quantitative R2*-based estimation of LIC at both 1.5T and 3.0T. IronQuant is an extension of the commercially available quantitative chemical-shift encoded (CSE) MRI method IDEALIQ (GE HealthCare, Waukesha, WI). The acquisition consists of a 3D multi-echo GRE sequence and confounder-corrected R2* parameter maps are then estimated by a joint nonlinear least squares optimization.^47,49–51 LIC is then calculated from the estimated R2* map according to a pre-calibrated linear relationship and returned inline.²⁹ Figure 5 shows example R2* and LIC maps from three clinical patients with varied levels of iron concentration.³⁰

Figure 5.

IronQuant, an advanced technical software module (ATSM) from GE HealthCare, calculates liver iron concentration (LIC) from a 3D confounder corrected R2* map. Estimated by nonlinear least squares from multi-echo spoiled gradient recalled echo (SPGR) data, IronQuant is compatible with both 1.5T and 3.0T acquisitions and allows full liver coverage within the duration of a single breath-hold (typically less than 20 seconds). On the left, the figure shows R2* and corresponding LIC maps from 3 patients with varied levels of iron concentration (female/36-yr/normal iron, male/23-yr/mild iron overload, and male/61-yr/severe iron overload) acquired during routine clinical evaluations. On the right, a slice of the LIC map from the patient with mild iron overload is shown as it would be displayed after the inline reconstruction. Imaging data courtesy Diego Hernando, PhD and Scott Reeder, MD, PhD, University of Wisconsin – Madison, acquired on 1.5T Signa Artist and 3.0T Signa Architect (GE HealthCare, Waukesha, WI).

Philips

Philips offers T2*/R2* measurements with the StarQuant and mDIXON Quant packages on all its systems. StarQuant uses a multi-echo GRE 2D or 3D Cartesian readout or a 3D radial stack-of-stars readout, and a mono-exponential fit is performed to realize the R2* maps. mDIXON Quant contains a 3D multi-echo GRE Cartesian readout and a mono-exponential fit, but also implements a 7-peak fat model to jointly reconstruct water, fat, PDFF in-phase, out-phase and B0 and fit error maps along with the T2*/R2* maps.^52,53 Both mDIXON Quant and StarQuant are compatible with SENSE parallel imaging, Compressed SENSE, and SmartSpeed reconstruction to reduce the acquisition time, which is especially important to minimize breath hold times.⁵⁴ To avoid breath holding, respiratory triggering or radial acquisition⁵⁵ can be used, which are works-in-progress for mDIXON Quant. Default StarQuant and mDIXON Quant protocols for T2*/R2* measurements are provided in Table 3. After scanning, measurement analysis can be performed on the IntelliSpace Portal (MR Liver Health) on whole organ, Couinaud segments or user-defined ROIs. The MR Liver Health package is being validated for LIC.⁶⁰ Figure 6 provides examples of the default 1.5T mDIXON Quant scan.

Figure 6.

IntelliSpace Portal MR Liver Health example of a default 1.5T mDIXON Quant scan including segmentation and histogram analysis with (a) T2* map, (b) PDFF map of the same slice, (c) report of the liver volume (1912.6 cm³), average PDFF (3.4 ± 6.4 %), average T2* (22.1 ± 7.5 ms) and average R2* (50.3 ± 18.7 s⁻¹) values with respective standard deviations, and (d) the liver volume T2* histogram.

Siemens Healthineers

The Siemens Healthineers “LiverLab“ product option consists of several sequences and convenience features. Simultaneous liver iron and fat evaluation is based on the 3D multi-gradient-echo (VIBE (volume interpolated breath-hold examination)) Dixon sequence. Scan assistance software optionally helps with volume positioning and field-of-view adaptation. Multi-echo data are processed inline by a multi-step adaptive pixel-wise fitting approach,⁵⁶ the inner core of which uses a confounder-correcting signal model containing water, fat, and R2*, with nonlinear fitting based on magnitude data; an initial complex-valued water-fat separation stage allows for a full dynamic range of the fat fraction between 0 and 100%. A model consistency metric called “GoodnessOfFit” map assists the user in performing quality assurance on each scan: its values, relative sum-of-squares residuals between fitted signal model and measured data, should be below 5%, indicating good agreement between fitted signal model and measured data.⁵⁷

Some inline reporting is integrated: a typical clinical measurement program always contains a high-resolution T1-weighted two-echo VIBE Dixon protocol as anatomical overview and/or pre-contrast-injection scan. For this, an inline liver segmentation and dual-ratio discrimination analysis can be activated.⁵⁸ The former yields a liver volume; the latter an early assessment whether iron overload, fatty infiltration, or a combination might be present. The liver segmentation result, together with an ROI which can be pre-planned at scan time is propagated to the processing of any subsequent multi-echo acquisition, where it is used for automated calculation and reporting of mean values, as shown in Figure 7. The maps can still be used for manual ROI analysis.

Figure 7.

Inline results from LiverLab product (Siemens Healthineers) at 3T of a 55-yr old male. From preceding two-echo Dixon a) Dual-ratio discrimination result (fat deposition), b) segmentation volume (1554 ml); from multi-echo Dixon c) report with extracted values of R2* (inline ROI 45s⁻¹ ~ LIC 0.6 mg/g using Hernando²⁹), PDFF (inline ROI 10.5%), and GoodnessOfFit (inline ROI 1.1% = very good) d) R2* map with projected liver segmentation, inline ROI and manual ROI (55s⁻¹, also showing the heterogeneity of the R2* distribution), e) corresponding PDFF map (manual ROI grayscale value 133 ~ 13.3%), f) cut-out of GoodnessOfFit map (manual ROI grayscale value 15 ~ 1.5% = very good).

With respect to iron overload, the product implementation was found to be linear up to at least ~1500 s⁻¹ and ~1600 s⁻¹ R2* at 1.5 T and 3 T, respectively with an optimized acquisition protocol.⁵⁹ Research versions (“works-in-progress” packages, WIPs) implement the creation of iron-unit LIC maps and reports with configurable calibration (Figure 8), explicit magnitude noise bias correction, higher tolerance against fat-water swaps, and 3D stack-of-radial free-breathing motion-corrected fat and iron evaluation.^38,55

Figure 8.

Inline results from multi-echo WIP at 1.5T of a 26-yr old male. a) LIC map with manual ROI (gray value 586 ~ 5.9 mg/g), b) report with extracted LIC values (inline ROI 5.7 mg/g), c) histogram of LIC values in segmentation region.

Offline Software

R2*-LIC

Various offline commercial and free software are available to estimate LIC from acquired multi-echo GRE images that are listed in Table 3. The only FDA-cleared commercial software among them that generates R2*-LIC reports is LiverMultiscan (Perspectum Diagnostics, Oxford, UK).⁶⁰ MRQuantif and parametricMRI are free software for R2*-based LIC estimation that provide the option for the user to select different signal models, ROIs and R2*-LIC calibrations and both software generate a R2*-LIC report.^20,61

R2* Parametric Maps

R2* maps can be generated offline from 2D or 3D multi-echo GRE images using various commercial and free software options. QLiver and MRmap are software primarily designed for producing liver R2* maps.^20,62,63 There are also other offline R2* software (QMass, CVI42, CMRtools) that are designed to estimate T2* for cardiac applications but they can be used to estimate liver R2* as well.^32,64,65 Lastly, in-house custom calculation using, e.g., Excel (Microsoft Corp., Redmond, WA, USA), MATLAB (The MathWorks, Inc., Natick, MA, USA) and other programming platforms can be performed to estimate both R2* and LIC, as they provide flexibility to the user for choosing signal models and R2*-LIC calibrations.^37,64,65

Accuracy and Reproducibility

Multiple studies have demonstrated the accuracy and robustness of MRI techniques to quantify LIC by showing excellent correlation with biopsy.^16,27,31,40 Analysis of diagnostic accuracy of LIC based on R2* with reference to Ferriscan-based LIC and biopsy based LIC demonstrated areas under the ROC curve greater than 0.95 and 0.80, respectively, which represents excellent accuracy of R2*-based LIC measurements.^29,62,73,74 Further, recent studies have demonstrated multicenter, multivendor reproducibility of R2/R2* based LIC quantification using similar acquisition parameters and a common processing algorithm or analysis software.^27,29,72 In summary, studies have shown that if there is a consensus on the image acquisition parameters and fitting algorithm then MRI based LIC measurements demonstrate high reproducibility.

Some important factors reducing the accuracy of MRI-based iron measurements are high iron concentration, low SNR, inhomogeneous main magnetic field (B0) and motion.⁷⁵ Using shorter or ultra-short echo times will extend the upper limit of R2* estimation and enable quantification of severe iron overload as discussed in the emerging and advanced MRI techniques described below. In cases of low SNR such as in high iron overload conditions, SNR can be enhanced by increasing the flip angle, which will improve the accuracy of R2* estimation. However, increasing the flip angle can cause T1 bias and overestimation of PDFF when a joint fat-R2* signal model is used.⁷⁶ For R2* methods, 3D GRE acquisitions are preferred over 2D because of the susceptibility to through-plane dephasing for 2D acquisitions that can cause bias in R2* measurements.⁷⁷ Another factor that can significantly bias MRI-LIC estimation is motion; nevertheless, several motion robust techniques have been proposed in recent years, which are described below in the emerging techniques.

To assess the reliability of the estimated R2/R2* maps, quality assurance metrics can be used. One vendor (LiverLab, Siemens Healthineers) provides a goodness-of-fit map for quality control, shown in Figure 7.⁷⁸ Confidence maps, e.g., based on signal model consistency and SNR can provide automated quality assurance of the estimated R2/R2* values and enable reliable ROI selection, which can improve the accuracy and reproducibility of MRI-based LIC estimation. A recent study has developed a fully automated algorithm to generate confidence maps and identify reliable regions for mean liver R2* estimation by accounting for the quality of acquired signal, B0 field map estimation, and local susceptibility effects.⁷⁵ These confidence maps can increase the reliability of MRI-based LIC estimation.

Another factor that can considerably affect accuracy and reproducibility in MRI-LIC measurements is the ROI selection. Previous studies have shown the usability of either drawing a whole liver ROI while excluding blood vessels or placing 3–4 small ROIs for LIC estimation.^79–82 However, the whole-liver ROI approach or maximizing the liver ROI area demonstrated higher reproducibility and inter-reader agreement compared to small ROI analysis.^66,83 Another advantage of the whole-liver approach is that it can enable the implementation of automated segmentation methods for liver extraction and inline LIC reporting.^84–86

Overall, to report accurate and reliable MRI-based LIC measurements, the technologists/radiologists should review the source images and quality maps such as the confidence maps to make sure there are no fat-water swaps or other artifacts, the liver signal on the first two echoes is above the noise floor and draw ROIs in homogeneous regions for accurate LIC estimation.^30,87 Further, for patient follow-ups it is advisable to use the same MRI method, acquisition protocol and calibration and ROI placement to obtain consistent LIC results and avoid any bias due to inter-method variability.⁸⁷

EMERGING AND ADVANCED TECHNIQUES

Motion Robust Techniques

Respiratory motion is a major challenge in abdominal imaging especially for children and patients unable to breath-hold. Some common motion-compensation strategies used are signal averaging and respiratory triggering. However, these techniques increase the imaging time and can still result in substantial motion artifacts and produce inaccurate iron quantification.^88,89 Recent studies have demonstrated the application of non-Cartesian k-space filling techniques such as radial GRE acquisitions to accurately estimate R2* in sedated children and patients unable to do breath-holds as they are less sensitive to motion in comparison to conventional Cartesian-based GRE acquisitions.^38,90 Moreover, radial acquisition can enable additional self-gated motion compensation, which restricts processing of signal dephasing arising from the combination of active imaging gradients and motion to ensure more precise liver R2* mapping.^55,91 Although non-Cartesian sampling strategies demonstrate improved accuracy in R2* measurements for free-breathing and pediatric applications, MRI vendors currently support these radial GRE and self-gating techniques as works in progress sequences for research only purposes (GE HealthCare, mDIXON, Philips and LiverLab, Siemens Healthineers^55,91,92), and more validation is needed before they can be clinically used for LIC assessments.

Ultra-short echo time (UTE) Imaging

Conventional GRE techniques used for R2* estimation have a minimum possible TE of ~1 ms and cannot capture the rapid MRI signal decay in tissues with severe iron overload at 1.5T (LIC > 25 mg Fe/g).^22,37 R2* estimation becomes even more restrictive at 3T for LIC > 12.5 mg Fe/g due to almost double the signal decay rate than at 1.5T.^12,93 To overcome this problem, MRI techniques with ultra short echo time (UTE), TE₁ ~ 0.1 ms have been proposed to capture the fast decaying signal.⁹⁴ Another advantage of UTE is that it uses radial acquisition and thus it is more robust to motion than conventional Cartesian GRE and is suitable for children and breath-hold noncompliant patients.⁹⁵ However, UTE imaging is susceptible to streaking artifacts and sensitive to system imperfections and gradient delays leading to out-of-slice signal contributions. Krafft et al. have implemented strategies to reduce streaking artifacts and suppress out-of-slice signals for estimating R2* using a multi-echo UTE acquisition and demonstrated excellent agreement between R2* and biopsy LIC for massive iron overload at 1.5 T and 3 T.⁹³ Further, Doyle et al. demonstrated that multi-echo UTE acquisition at 3 T increased the dynamic range of LIC quantification up to 50 mg Fe/g which exceeded the 1.5 T GRE achievable LIC limits.⁹⁶ In addition, simulation, phantom and patient studies have shown superior accuracy and precision for multi-echo UTE acquisition for the entire physiological LIC range compared to conventional GRE acquisition.^95,97,98 A further improvement in UTE acquisition has been shown by Kee et al. by using a 3D cone trajectory rather than radial trajectory, which increased SNR efficiency and demonstrated improved R2* estimation and image quality comparable to breath-hold GRE in children.^99,100 Despite the feasibility of UTE in expanding the dynamic range of clinical LIC, UTE imaging has not been clinically adopted yet due to the lack of a thorough biopsy-referenced or R2-referenced clinical study showing the agreement between UTE R2* and LIC measurements, lack of availability of multi-echo UTE sequences from the vendor, and the possibility of picking up signals from the short-T2* component in liver that may confound R2* and LIC estimation.¹⁰¹

Quantitative Susceptibility Mapping (QSM)

QSM is a novel alternative technique for direct measurement of iron concentration by quantifying the underlying magnetic susceptibility of tissue.¹⁰² Further, it is unaffected by cellularity changes such as fibrosis and provides susceptibility values independent of field strengths, while relaxometry measurements are dependent on field strength.^102–104 QSM techniques are well validated to measure iron deposits in the brain. However, abdominal QSM can face technical challenges such as respiratory motion, presence of fat and rapid signal decay in case of severe iron overload (<2 mg Fe/g in brain vs. >10 mg Fe/g in liver) that can hinder accurate quantification of susceptibility maps.¹⁰² Over the last decade, QSM has been evaluated with conventional multi-echo 3D GRE techniques which are employed for both R2* and PDFF estimation.^105–108 Studies have shown moderate-to-strong correlation between LIC estimated by QSM and relaxometry (R2, R2*).^102,103,108 A recent effort has shown the implementation of a motion-resolved image reconstruction using a free-breathing 3D multi-echo UTE cones acquisition for liver QSM estimation, which demonstrated improved image resolution, significant reduction in motion artifacts, and estimation of QSM values similar to breath-hold Cartesian GRE acquisition.¹⁰⁴ Despite the promise of QSM as a biomarker for LIC, it is still limited to research due to the unavailability of a commercial QSM processing software for clinical routine, susceptibility values being relative measurements with respect to a reference tissue, lack of a fully characterized conversion factor from susceptibility to LIC, and insufficient validation in the presence of steatosis.^102,109

MRI-LIC FOR CLINICAL DECISION-MAKING

Accurate assessment of LIC is critical for guiding iron chelation therapy to reduce iron overload complications while maintaining normal LIC ranges to avoid chelator toxicity.^4,5 MRI techniques have not only enabled non-invasive and accurate estimation of LIC but also improved tailoring of iron chelation therapy and making important and timely clinical decisions for patients with iron overload.^110–112 Further, MRI allows close longitudinal monitoring of LIC facilitating titration of chelation therapy.^110,111

Of all the existing MRI techniques, the Ferriscan R2 method is the most accepted clinically, because it was the earliest methods to gain regulatory approval for LIC reporting.^30,87 Further, FerriScan takes the sole responsibility of quality assurance, data processing and report generation whereas for R2* method the regulatory approval is not yet available, and the responsibility shifts onto interpreting radiologists and/or clinicians and their institution for R2*-based LIC estimation. However, the R2* method has much to offer in terms of reducing temporal and monetary costs while demonstrating high reproducibility as shown in a recent multi-center, multi-vendor study.²⁹ Further, some recent studies demonstrated that between R2 and R2* methods, there is no significant difference between the rate of change of LIC over serial examinations and in estimates of iron chelation efficacy, hence signifying either R2 or R2* method can be used in individual patients for guiding chelation therapy.^70,113 However, both studies concluded that R2-LIC and R2*-LIC are not directly interchangeable, and the same method should be used longitudinally when possible in clinical practice.^70,113,114 The discrepancies between R2 and R2* measurements can be due to their differential sensitivity to the pattern and scale of iron deposition arising from patient-specific differences in iron distribution and clearing mechanisms.^12,113,115 In contrast, another group of investigators demonstrated good agreement between Ferriscan and R2*-LIC-derived clinical decisions and showed that switching from Ferriscan-LIC to R2*- LIC estimation impacted the clinical decision-making similar to switching from one hematologist to another.⁷³

Apart from LIC, monitoring iron burden in other organs including the heart and pancreas is important clinically for diagnosing diseases associated with iron overload and guiding chelation therapy and risk-stratifications in some patients.^4,36 R2* MRI techniques provide a comprehensive non-invasive assessment of iron in several extrahepatic organs, of which cardiac T2* is the most common measure used clinically to avoid cardiac complications related to iron overload.^43,116–118 Although R2*-iron calibrations are not used or derived for heart and pancreas unlike R2*-LIC, T2*/R2* values are still important measures used by clinicians in these organs for risk assessment and guiding iron chelation therapy.^117–119 Hence, MRI techniques have become essential and critical for assessment of iron overload and guiding treatment.

OUTLOOK

Non-invasive, MRI-based quantification of iron overload to guide patient management is becoming essential for clinical radiology practices. However, the variety of MRI techniques, each with different processing workflows, is adding complexity to its widespread application for clinical care. Both R2 and R2* MRI methods are accurate and reproducible for calculating LIC estimates in clinical practice when performed using validated acquisition parameters and analysis protocols. Emerging R2*-MRI techniques based on radial sampling and UTE imaging will improve motion compensation and increase the dynamic range of R2*-LIC estimation. In the past, the lack of standardization of acquisition parameters, signal model and calibration has posed a setback for R2* based LIC quantification. Recent technical consensus, multi-center studies and ESGAR/SAR guidelines demonstrate that R2* MRI is the most accurate and reproducible method for LIC estimation and the ongoing efforts in vendor implementation will soon make R2*-MRI the standard-of-care for the assessment of iron overload.