Quantitative Susceptibility Mapping in Combination with Water-Fat Separation for Simultaneous Liver Iron and Fat Fraction Quantification

Abstract

Purposes

To evaluate the feasibility of simultaneous quantification of liver iron concentration (LIC) and Fat Fraction (FF) using water-fat separation and quantitative susceptibility mapping (QSM).

Methods

Forty-five patients suspected of liver iron overload (LIO) were included. A volumetric interpolated breath-hold examination sequence for QSM and FF, a fat-saturated gradient echo sequence for R₂^*, a spin echo sequence for LIC measurements and MRS analyses for FF (FF-MRS) were performed. Magnetic susceptibility and FF were calculated using a water-fat separation method (FF-MRI). Correlation and Receiver operating characteristic analyses were performed.

Results

Magnetic susceptibility showed strong correlation with LIC (r_s=0.918). The optimal susceptibility cutoff values were 0.34, 0.63, 1.29, and 2.22 ppm corresponding to LIC thresholds of 1.8, 3.2, 7.0, and 15.0 mg/g dry weight. The area under the curve (AUC) were 0.948, 0.970, 1, and 1, respectively. No difference in AUC was found between susceptibility and R₂^* at all LIC thresholds. Correlation was found between FF-MRI and FF-MRS (R²=0.910).

Conclusions

QSM has a high diagnostic performance for LIC quantification, similar to that of R₂^*. FF-MRI provides simultaneous fat quantification. Findings suggest QSM in combination with water-fat separation has potential value for evaluating LIO, especially in cases with coexisting steatosis.

Introduction

Iron overload syndrome is a common clinical problem resulting from diseases of iron hyperabsorption, as well as transfusion therapy [1]. Since no active excretion mechanism exists for excess iron of the body, iron deposes in major organs and thus may cause complications, such as liver disease and cardiomyopathy [2]. A regular monitor and clinical intervention, when necessary, are critical to prevent complications of iron overload [3]. Therefore, accurate total iron assessment is essential for the diagnosis of iron overload and for treatment monitoring to maintain relatively low iron deposits while minimizing side-effects [1].

Liver iron concentration (LIC) is considered as a reliable marker of total body iron status [1]. MRI-based R₂ and R₂^* mappings are two commonly used approaches for LIC quantification. Image-based FerriScan-R₂ (Resonance Health, Claremont, Australia) is a regulatory-approved standardized spin-echo measurement with quality-controlled iron reporting. However, high price and long scanning time narrow its widespread use [4]. Previous studies have shown that R₂^*-based MRI iron quantification approaches have strong correlations with LIC [5; 6]. However, the ability to evaluate iron level through R₂^* is influenced by the presence of fat [7]. Commonly, fat and iron coexist in patients with diffuse disease [8–11] (e.g. hereditary hemochromatosis [10] and nonalchoholic steatohepatitis [11]). The presence of fat causes additional signal modulations with echo time that lead to a positive bias in R₂^* estimation [12]. Therefore, fat-saturation techniques were advocated by some groups [13; 14] to minimize fat signal contributions. Fat saturation could be beneficial in patients with high lipid content and relatively low iron accumulation in the liver [14]. However, it also suppresses part of the water signal, which leads to lower signal-to-noise ratio (SNR) and bias R₂^* quantification [15; 16]. In addition, coexisting steatosis in patients with iron overload is a relevant cofactor associated with the accelerated disease progression [17]. Simultaneous quantification of fat and iron overload benefits clinical management. Thus, several studies have described methods to simultaneously quantify Fat Fraction (FF) and R₂^* relaxation time in the liver [7; 11; 18–20]. In the work by Henninger et al. [20], a 3D-multiecho-Dixon (3D-ME-Dixon) was utilized for the simultaneous quantification. However, fat/water swaps are still problematic for Dixon techniques for patients with high LIC due to the low SNR of the original images [20].

Quantitative susceptibility mapping (QSM) is a relatively novel approach to evaluate iron deposition in vivo which is independent of imaging parameters. QSM utilizes the measured B₀ field to quantify the magnetic properties of tissue, such as paramagnetic compounds like iron in the cell body [21–24]. However, the application of QSM in the liver faces an additional challenge not critical in the brain - the chemical shift caused by fatty tissue. This chemical shift affects the complex-valued MRI signal, particularly the phase signal, and further QSM quantification. In this study, the confounding chemical shift was removed using a water-fat separation method [25]. Magnetic susceptibility maps were reconstructed based on the estimated B₀ field maps, and FFs were obtained from the resultant water and fat components (FF-MRI). Therefore, iron and fat level could be evaluated simultaneously from susceptibility maps and FF calculation.

The primary purpose of this study is to evaluate the diagnostic performance of QSM in liver iron quantification, with FerriScan LIC as the reference, and compare it with that of R₂^* derived from a fat-saturated multi-echo gradient echo (GRE) sequence. Furthermore, the accuracy of FF-MRI was explored by comparing with MRS-based T₂-corrected FF (FF-MRS) [26].

Methods

Patients

This prospective study was approved by the institutional review board. An informed consent was submitted by each patient. Patients suspected of liver iron overload with an elevated serum ferritin level (>500 ng/mL) were included in the study. From July 2015 to December 2016, a consecutive series of 61 (40 men, 21 women) potential study participants were approached for inclusion in this study, and 59 consented to participate. Of them, 3 patients were excluded due to claustrophobia. Data of 11 patients were unsuitable for further analysis due to obvious lesions in the right lobe of liver (n= 2) or extremely massive liver iron overload (n=9) based on MRI scans. These 9 patients with FerriScan-LIC greater than 27.5 mg/g dw showed extremely fast MRI signal decay and no reliable signal can be used for accurate field map estimation. Therefore, 45 patients with successful MRI scanning and suitable data for analysis were included in this study (Fig. 1). Demographic and clinical characteristics are shown in Table 1.

Figure 1.

Flowchart of the patient inclusion process.

Table 1.

Summary of Demographic and Clinical Characteristics

Parameter	Value	Parameter	Value
Demographic characteristics		Clinical characteristics
No. of patients	45	Myelodysplastic syndrome	11
Sex		Aplastic anemia	9
No. of men	31	Chronic liver disease	5
No. of women	14	Thalassemia	3
Mean age(y)*		Hemochromatosis	2
All patients	44.36 (18~73)	leukemia	2
		pancytopenia	1
Men	44.45 (18~71)	Unknown	12
		Serum ferritin level (ng/ml)
Women	44.14 (18~73)	Mean ± SD†	1337.65±717.35
		Range	512.8~3947.4

^*Numbers in parentheses are ranges;

^†SD=standard deviation.

Magnetic Resonance Imaging

All MRI examinations were performed on a 1.5T scanner (MAGNETOM Aera; Siemens Healthcare, Erlangen, Germany) using a 6-channel body matrix coil and an integrated spine matrix coil. A prototype multi-echo volumetric interpolated breath-hold examination (VIBE) sequence was acquired for QSM and FF in a single breathhold of 19s, and a multi-echo 2D GRE sequence with fat saturation was performed for R₂^* in 28s with 2 breathholds (14s for each). In addition, a spin echo sequence for FerriScan-R₂ data acquisition were performed according to the manual of FerriScan centre. Table 2 lists the relevant parameters for each of the three sequences.

Table 2.

Acquisition Parameters for Sequences

Sequence	TE1 (ms)	Echo spacing	Number of echoes	TR (ms)	Flip angle (deg)	Bandwidth (Hz/pixel)	FOV (mm2)	Matrix size	Slice thickness (mm)	Number of slice
3D VIBE	1.44	1.36	6	10	6	1040	420×315	224×168	5	44
2DGRE (fat-saturated)	1.09	1.56	12	50	20	1500	400×275	128×88	10	10
2D FerriScan	6	3	5	1000	90	500	400×300	256×192	5	11

Magnetic Resonance Spectroscopy

The prototype MRS sequence called HISTO [26] for the reference FF estimation was scanned with the following parameters: TR=3000 ms; TE₁~TE₅=12/24/36/48/72 ms; averages=1; bandwidth=1200 Hz; vector size=1024; voxel size=12~27 cm³; acquisition duration=15 s in one breath-hold. The volume of interest was placed on the right lobe of the liver, avoiding the main vessels and liver edges as possible.

QSM Reconstruction and FF calculation

QSM images were reconstructed using STI Suite software (https://people.eecs.berkeley.edu/~chunlei.liu/software.html). Briefly, the phase offsets between odd and even echoes are corrected prior to water-fat separation reconstruction and incorporated with a predefined fat spectrum model with 6 peaks, at [−244.3, −221.7, −175.4, −119.3, −32.1, 34.0] Hz, with relative amplitudes 0.01 · [9.45e^−iπ0.181, 64.66, 9.67e^iπ0.046, 2.26e^−iπ0.567, 2.22e^−iπ0.244, 8.83e^−iπ0.089] [25; 27]. Then, the field map is estimated using a water-fat separation method [25]. The estimated B₀ field is processed using V-SHARP to remove the background phase [28]. The filtered phase is further processed using a two-level STAR-QSM (streaking artefact reduction for QSM) algorithm [29; 30]. A schematic view of QSM reconstruction and FF calculation was shown in Fig. 2. More details can be found in the Supplementary material.

Figure 2.

Processing pipeline to estimate the susceptibility map and fat fraction map of liver. First, complex multi-echo images were used to estimate the B₀ field map, water, fat and R₂^* images (Fig. 2b). Second, the sum-of-squares of the magnitude images were used to obtain a mask (Fig. 2c). This binary mask provided edge information to calculate the local field by removing the background field (Fig. 2d). The local field map is subsequently input to the two-level regularization approach to obtain the QSM maps (Fig. 2e). FF is calculated from estimated water and fat images: FF = fat/ (water + fat) *100% (Fig. 2f).

Data Analysis

VIBE and GRE datasets were analysed by two radiologists specializing in liver imaging (reader 1: HM. L with 3 years of experience, and reader 2: NY. H, with 5 years of experience). R₂^* was computed with a three-parameter curve-fitting model (offset model) [31] on a workstation (ADW 4.6, GE Healthcare). Susceptibility values and FF-MRI were measured with ImageJ software (version 1.6; National Institutes of Health, Bethesda, MD). The acquired SE MR data were subsequently uploaded to Resonance Health (http://www.resonancehealth.com/) for FerriScan- LIC assessments.

On susceptibility maps, circular regions of interest (ROIs) were drawn in the inferior slices of right hepatic lobe, avoiding vessels, borders, and artefacts. Liver susceptibility values were referenced to mean susceptibility value of the paravertebral muscle tissue by placing additional ROIs at the renal hilum level, since paravertebral muscle doesn’t accumulate excess iron [21]. ROIs on susceptibility images were drawn on three consecutive slices (Fig. 3). The ROI on the R₂^* map was delineated in a visual alignment with the liver ROI of the susceptibility map, and the ROI of FF-MRI was approximately co-localized to the MRS voxel.

Figure 3.

Susceptibility value, R₂^* and R₂ measurement in a 37-year-old man with elevated Serum Ferritin of unknown aetiology. (a) Top, the liver susceptibility value was measured by placing the circular ROI in segment VI. Bottom, another reference ROI was placed in the paravertebral muscle tissue at the renal hilum level. (b) Top, the ROI of R₂^* was drawn in a visual assignment with that of QSM image in (a). Bottom, signal decays with increasing TE. (c) Top, R₂ map shows that large vascular structures and other image artefacts were excluded from the analysis of R₂ map. Bottom, the R₂ histogram illustrates the R₂ estimation of 51.1±12.2 s ⁻¹. The LIC was 2.0 mg/g dw (normal range, 0.17~1.8 mg iron per gram dry tissue).

The spectra data was processed using the method described in [26] to estimate the T₂-corrected FF. The integrals of water and fat signal on the real part of the spectra at each TE were automatically generated by the syngo Spectroscopy software (Siemens Healthcare). Manual adjustment in phase and baseline correction steps was executed in case that the automatic scheme failed. Subsequently, the integrals were fit to a monoexponential model using an in-house developed MATLAB (Mathworks, Natick MA) program to obtain the T₂-corrected water and fat content (S₀_fat and S₀_water). Finally, the T₂-corrected fat fraction was calculated as FF = S₀_fat / (S₀_fat + S₀_water) *100%.

Statistical Analysis

Intraclass correlation coefficients (ICC) were performed to test the inter-rater reliability between measurements (QSM, R₂^*) by two readers. An ICC of 0.8–1.00 was considered to indicate excellent agreement; 0.61–0.8, good agreement; 0.41–0.60, moderate agreement; 0.21–0.40, fair agreement; and 0.2 or less, slight agreement. If excellent agreement was obtained, the datasets were averaged for further analyses. The data were descriptively analysed and statistically tested for normality using the Shapiro-WiIk test. Spearman correlation analyses were performed to analyse the correlations between QSM and FerriScan-LIC, and between R₂^* and FerriScan-LIC. Linear regression analysis was performed to compare the FF-MRI and FF-MRS.

Five grades (normal, mild, moderate, heavy, and extremely-heavy grade) were assigned using the clinical LIC grading thresholds (1.8, 3.2, 7.0, and 15.0 mg/g dw). FerriScan-LIC was set as the reference for LIC grades. Receiver operating characteristic (ROC) analysis was performed and the area under the ROC curve (AUC) was calculated for both QSM and R₂^*. The optimal cutoff value was the value at which the sum of the sensitivity and specificity was maximized. Two paired ROC curves were compared using the method developed by DeLong et al. [32].

All these statistical analyses were performed using the SPSS software (SPSS version 23.0; SPSS Inc., Chicago, IL) and MedCalc software (MedCalc version 15, Mariakerke, Belgium). The statistical significance threshold was set at P<0.05.

Results

Susceptibility and R₂^* Assessment

All 45 patients were included for the estimation of susceptibility, R₂^*, and LIC. The mean size of the liver and muscle ROIs on the susceptibility map was 1.72±0.46 cm² (range, 1.04~2.59 cm²) and 0.79±0.28 cm² (range, 0.39~1.23 cm²), respectively. The ICCs for susceptibility and R₂^* between the two observers were 0.973 and 0.998, respectively. The mean value for susceptibility, R₂^* and LIC were 1.25±1.41 ppm (range, -0.23~5.94 ppm), 236.14±247.47 s⁻¹ (range, 32.98~926.49 s⁻¹), and 6.70±7.43 mg/g dw (range, 0.5~27.5 mg/g dw), respectively. Shapiro-WiIk test showed a non-normal distribution for all these datasets.

Correlations between Susceptibility and FerriScan-LIC, and between R₂^* and FerriScan-LIC

With increased iron concentration confirmed by FerriScan-LIC, VIBE magnitude images showed consistent signal decay, and both R₂^* and susceptibility values were increased accordingly (Fig. 4). Positive correlations were found (Fig. 5) between susceptibility and FerriScan-LIC (r_s=0.918; 95% confidence interval [CI]: 0.855, 0.954; P<0.001), and between R₂^* and FerriScan-LIC (r_s=0.946; 95% CI: 0.903, 0.970; P<0.001).

Figure 4.

Magnitude (original VIBE images at TE=8.24 ms), R₂^*, QSM maps and FerriScan-R₂ for three subjects, including 18-year-old man with leukaemia, 18-year-old woman with aplastic anaemia, and 59-year-old woman with aplastic anaemia.

Figure 5.

Scatterplot shows a positive correlation between QSM and FerriScan-LIC, and between R₂^* and FerriScan-LIC (r_s=0.918, P<.001; r_s=0.946, P<.001, respectively).45 patients were included for the estimation of QSM and R₂^*.

Diagnostic Performance of Grading based on LIC levels

The optimal susceptibility cutoff values for each LIC threshold were 0.34 ppm (1.8 mg/g dw), 0.63 ppm (3.2 mg/g dw), 1.29 ppm (7.0 mg/g dw), and 2.22 ppm (15.0 mg/g dw),respectively. For R₂^*, the optimal cutoff values were 58.01, 158.86, 282.94, and 469.31 s⁻¹, respectively. The AUCs for QSM cutoff values were 0.948, 0.970, 1 and 1, while for R₂^* thresholds were 0.969, 0.972, 0.997, and 1, respectively. There was no significant difference between the AUCs for QSM and R₂^* at all the four LIC grading levels (Table 3).

Table 3.

Comparison of LIC Grading Performance with QSM and R₂^*.

LIC Threshold (mg/g dw)	Susceptibility value (ppm)				R2* (s−1)				Z Value	P Value
	Optimal cutoff(ppm)	Sensitivity	Specificity	AUC	Optimal cutoff (s−1)	Sensitivity	Specificity	AUC
1.8	0.34	0.97 [35/36] (0.86,0.99)	0.78 [7/9] (0.4,0.97)	0.948 (0.837,0.992)	58.01	1 [36/36] (0.93,1)	0.89 [8/9] (0.52,1)	0.969 (0.869,0.998)	0.57	0.57
3.2	0.63	0.96 [21/22] (0.77,1)	0.96 [22/23] (0.78,1)	0.970 (0.871,0.998)	158.86	0.91 [19/21] (0.71,0.99)	1 [23/23] (0.85,1)	0.972 (0.874,0.999)	0.06	0.95
7	1.29	1 [12/12] (0.735,1)	0.97 [32/33] (0.84,1)	1 (0.916,1)	282.94	1 [12/12] (0.74,1)	0.97 [32/33] (0.84,1)	0.997 (0.916,1)	0	1
15	2.22	1 [7/7] (0.59,1)	1 [38/38] (0.91,1)	1 (0.921,1)	469.31	1 [7/7] (0.59,1)	1 [38/38] (0.91,1)	1 (0.921,1)	0	1

Note: Numbers in brackets are raw data. Numbers in parentheses are 95% confidence intervals. FerriScan-LIC was used as the standard of reference.

FF Assessments with MRI and MRS

The fat peaks of 20 patients were difficult to be recognized from MRS, probably due to iron-related rapid signal decay (n=10, LIC ranging from 9.9 to 27.5 mg/g dw) and extremely low-fat content (n=9). Besides, 2 cases were also excluded because of both the high iron (LIC=8.8 and 21.3 mg/g dw, respectively) and extremely low-fat content. Thus, the remaining 24 patients were included in the FF analysis. The mean FF-MRI and mean FF-MRS were 8.00% (range, 2.00%~22.90%), and 5.69% (range, 0.41%~18.87%). FF-MRI showed a strong correlation with FF-MRS (R²=0.910, P<0.01). The linear regression line (Fig. 6) between FF-MRI and FF-MRS had a slope of 1.09 (P<0.01) and an intercept of 1.91(P<0.01).

Figure 6.

Correlation between FF-MRI and FF-MRS for 24 patients (R²=0.910, P<.001).

Discussion

The findings of this study demonstrated that QSM showed strong correlation with FerriScan-LIC (r_s=0.918). ROC analysis further demonstrated that QSM could grade LIC (normal to extremely-heavy grades), with high diagnostic performance similar to that of R₂^*. Furthermore, FF-MRI was strongly correlated with FF-MRS (R²=0.910). These results suggest that QSM provides the necessary information to assess LIC, and that fat fraction can be quantified from the same single scan using a water-fat separation method.

In this study, the strong correlation (r_s= 0.918) between QSM and FerriScan-LIC was consistent with a previous study [33] (r_p=0.872 at 1.5T), indicating the role of QSM as a promising imaging biomarker for quantifying hepatic iron overload. This previous study set subcutaneous adipose tissue as the susceptibility reference [33], while we use paravertebral muscle as the susceptibility reference. High correlation coefficients for these two approaches demonstrate that either subcutaneous adipose tissue or paravertebral muscle can be used as the susceptibility reference. Some of the participants recruited in this study showed little accumulation of subcutaneous adipose tissue, which could be due to hypermetabolic conditions [34]. Limited adipose area increased the difficulty to draw the reference ROIs for reference susceptibility value calculation. Thus, as proposed by the previous study [21], we choose paravertebral muscle as the reference.

The value of R₂^*-based MRI techniques has been approved by the American Society of Haematology [35]. Although the R₂^* bias increased with R₂^*, studies demonstrated the bias induced by the use of saturation were small and did not cause a clinically significant difference, when the measured R₂^* values were confined to a range below 500 s⁻¹ [13; 15; 16]. In the current study, 6 of the 45 patients had R₂^* values above 500 s⁻¹. This could be the reason why the estimated R₂^* based on the fat-saturated GRE sequence showed good diagnostic performances, with all the AUCs greater than 0.95. ROC analyses demonstrated the high diagnostic performance of QSM to quantify LIC. Meanwhile, the diagnostic performance of QSM showed no significant difference with that of R₂^* at the LIC thresholds of 1.8, 3.2, 7.0, and 15 mg/g dw. LIC greater than 1.8 mg/g dw is considered to reach the diagnosis of iron overload. LIC greater than 3.2 mg/g dw is the clinically significant threshold for iron chelation therapy, the LIC of 7.0 mg/g dw is a key threshold for iron chelation therapy initiation and efficacy monitoring, and LIC of 15 mg/g dw is an indicator of substantial risk for cardiac diseases and early death in thalassemia [36]. Hence, QSM might be useful not only in the detection of lower levels of iron accumulation, but also in the guidance of iron chelation therapy.

Coexisted fat and iron play an interactive aggressive role in the disease progression [8; 9]. As we know, oxidative stress induced by iron accumulation within hepatocytes contributes to the end-stage cirrhosis and development of hepatocellular carcinoma in patients with hemochromatosis. Moreover, cirrhosis plays a major role for the increased risk of liver cancer [37]. Steatosis has shown to be a relevant cofactor adding to the effect of iron in accelerating fibrosis [17]. Steatosis also shows the predisposition to diabetes, which is considered as a clear risk factor for advanced hepatic fibrosis [10]. Management of cofactors is important to reduce the risk of liver fibrosis [10], consequently decreasing the risk of liver failure and liver cancer. Thus, simultaneous iron and fat quantification is imperative. Interestingly, FF-MRI from the water-fat separation method showed strong correlations with FF-MRS. Previous studies of hepatic MRS have demonstrated its value as a useful examination of lipid content [34]. Meanwhile, a previous study demonstrated that T₂-corrected FF by MRS was not correlated with the histologic degree of iron deposition [38]. However, MRS voxels are acquired in a breath-hold state according to the manual placement on its localizer images. There are still possible inconsistencies between the actual acquired voxel position and the planned position. Besides, FF-MRS reflects the FF of voxels, while FF-MRI is a ROI-based fat estimation [20]. Both of them might be the reasons for the bias between FF-MRS and FF-MRI, even though FF-MRS is accepted as a "gold standard" and visual alignment was performed as possible. Note that, water-fat separation reconstruction was conducted using a predefined fat spectrum model derived from food oil experiments [25; 27], in the present study. Different fat spectra derived from liver MRS [27], are also compatible for the water-fat separation in our study. Recently, one study by Hong et al. has demonstrated that there was no obvious bias for liver PDFF estimation when using different biologically plausible fat spectral models [39].

These findings demonstrated that QSM is feasible to evaluate hepatic iron, with simultaneous fat fraction estimation. Based on the water-fat separation model, we can estimate water, fat, R₂^* and B₀ field map. From these estimated components, the FF and susceptibility value are computed. The computed susceptibility values showed a strong correlation with FerriScan-LICs, which provides a new insight to quantify the iron content that coexists with fat. Although the manufacturer-provided MRS was chosen for the reference FF estimation in cases with coexisting iron by many studies [20; 40], the relative long TE_min (12 ms) and TE spacing was not applicable in cases with extremely high iron levels. While FF-MRI can provide an estimation of FF in these high-iron cases, the accuracy is not evaluated due to the lack of an FF-MRS reference. Nevertheless, FF-MRI showed a strong correlation with FF-MRS in lower LIC levels, which still demonstrated that water-fat separation is promising for fat quantification. This is, to our best knowledge, the first study reporting the simultaneous evaluation of iron and fat using QSM in combination with a water-fat separation method.

There are several approaches with respect to simultaneous fat and R₂^* measurements [7; 11; 18–20]. In the works of Galimberti et al. [18] and Franca et al.[19], 2D multi-echo GRE sequences were utilized to accurately quantify liver R₂^* and FF simultaneously for patients with iron overload. Henninger et al. [20] used a 3D-ME-Dixon for simultaneous iron and fat quantification. Iron and fat can be measured throughout the liver within a breath-hold with this sequence, while 2D GRE might need multiple breath-hold acquisitions at lower slice resolution. The 3D-ME-Dixon sequence was considered as a valuable tool for the estimation of hepatic iron and fat in a clinical setting. However, with the inline calculation used in their study, fat/water-swaps remained a drawback, and the measurable R₂^* values were constrained to 400 s⁻¹ in order to avoid deviation induced by the inline fitting algorithm of that particular version [41]. Note that the highest estimated R₂^* value depends on different parameters, for example, field strengths, imaging acquisition parameters , and different fitting algorithms [42]. In the present study, QSM analyses were available for patients whose R₂^* were up to 926.49 s⁻¹ using a similar 3D Dixon sequence, though QSM reconstruction needs a few steps off-line post-processing.

There were also several limitations in our study. First, we employed FerriScan-based LIC and FF-MRS as the references. On one hand, liver biopsy was not performed considering safety reasons. On the other hand, regulatory-approved FerriScan-LIC was verified as a reliable and accurate method to evaluate LIC in a multicentre validation study [43]. Although FF-MRI showed strong correlation with FF-MRI in lower LIC levels, the feasibility of FF-MRI estimation in cases with higher iron still need further studies. Second, the possible interferences of confounding factors (i.e. inflammation and fibrosis) with liver susceptibility measurements were not investigated in the present study. A previous study has shown that inflammation and fibrosis contribute diamagnetic (negative) susceptibility in kidney [44]. Interferences of inflammation and fibrosis with liver susceptibility need to be fully considered in further studies. Third, QSM was not available in the quantification of extremely massive iron overload (e.g. LIC>27.5 mg/g dw). For these patients with extremely iron overload, the signal decays faster, and phase values may be below the noise level, and phase unwrapping is challenging with low signal-to-noise ratio. Thus, B₀ field may be underestimated, which is a limitation of QSM for quantifying extremely iron overload. It could be addressed using ultrashort TE sequences in future studies [45]. As for R₂^*, extremely rapid signal decay may also impose a limitation for accurate iron quantification, though it can be addressed by shorter echo spaces and minimum first TE [46]. Considering that the susceptibility and R₂^* are based on different contrast mechanisms, although R₂^* shows a little bit higher correlation with LIC, the R₂^* suffers from non-local susceptibility effects and still confounds its accuracy for evaluating iron content [47]. Finally, using a flip angle of 6° at a TR of 10ms in current study, may have introduced the possibility of T₁ weighting in the FF estimation. Since the signal decays faster and higher SNR is critical for accurately estimating B₀ field map in cases with high iron levels, the flip angle of 6°, also used in one previous study [48], was chosen as a tradeoff between the T1 bias and adequate SNR for B₀ field map estimation.

In conclusion, QSM in combination with water-fat separation is valuable in quantification and grading of liver iron overload, especially in cases with coexisting steatosis.

Key Points.

• Magnetic susceptibility showed strong correlation with LIC (r_s=0.918).

• QSM showed high diagnostic performance for LIC, similar to that of R₂^*.

• Simultaneously estimated FF-MRI showed strong correlation with MR-Spectroscopy-based FF (R²=0.910).

• QSM combining water-fat separation has quantitative value for LIO with coexisted steatosis.

Abbreviations

AUC

the area under the curve

FF-MRI

FF calculated with a water-fat separation method

FF-MRS

MR Spectroscopy based FF

FF

Fat Fraction

FOV

field of view

GRE

gradient echo

ICC

Intraclass correlation coefficients

LIC

liver iron concentration

LIO

liver iron overload

mg/g dw

mg/g dry weight

QSM

Quantitative Susceptibility Mapping

ROC

Receiver operating characteristic

ROI

region of interest

VIBE

volumetric interpolated breath-hold examination