Validation of liver quantitative susceptibility mapping across imaging parameters at 1.5 T and 3.0 T using SQUID susceptometry as reference

Abstract

Purpose

To validate QSM‐based biomagnetic liver susceptometry (BLS) to measure liver iron overload at 1.5 T and 3.0 T using superconducting quantum interference devices (SQUID)‐based BLS as reference.

Methods

Subjects with known or suspected iron overload were recruited for QSM‐BLS at 1.5 T and 3.0 T using eight different protocols. SQUID‐BLS was also obtained in each subject to provide susceptibility reference. A recent QSM method based on data‐adaptive regularization was used to obtain susceptibility and ${\mathrm{R}}_2^{*}$ maps. Measurements of susceptibility and ${\mathrm{R}}_2^{*}$ were obtained in the right liver lobe. Linear mixed‐effects analysis was used to estimate the contribution of specific acquisition parameters to QSM‐BLS. Linear regression and Bland–Altman analyses were used to assess the relationship between QSM‐BLS and SQUID‐BLS/${\mathrm{R}}_2^{*}$.

Results

Susceptibility maps showed high subjective quality for each acquisition protocol across different iron levels. High linear correlation was observed between QSM‐BLS and SQUID‐BLS at 1.5 T (r ² range, [0.82, 0.84]) and 3.0 T (r ² range, [0.77, 0.85]) across different acquisition protocols. QSM‐BLS and ${\mathrm{R}}_2^{*}$ were highly correlated at both field strengths (r ² range at 1.5 T, [0.94, 0.99]; 3.0 T, [0.93, 0.99]). High correlation (r ² = 0.99) between 1.5 T and 3.0 T QSM‐BLS, with narrow reproducibility coefficients (range, [0.13, 0.21] ppm) were observed for each protocol.

Conclusion

This work evaluated the feasibility and performance of liver QSM‐BLS across iron levels and acquisition protocols at 1.5 T and 3.0 T. High correlation and reproducibility were observed between QSM‐BLS and SQUID‐BLS across protocols and field strengths. In summary, QSM‐BLS may enable reliable and reproducible quantification of liver iron concentration.

1. INTRODUCTION

Excessive accumulation of iron in the body is toxic and may result in multiple health complications including liver and heart damage, pancreatic dysfunction, and growth failure. Liver iron concentration (LIC) is widely recognized as the best surrogate measurement for the assessment of body iron content. ¹ Although LIC can be determined from percutaneous liver biopsy samples, biopsy is invasive, suffers from sampling variability, ² and is contraindicated in certain patients because of the risk of uncontrolled bleeding. Therefore, accurate and noninvasive LIC measurement is needed to evaluate LIC in patients with liver iron overload.

Superconducting quantum interference device (SQUID)‐based biomagnetic liver susceptometry (BLS) is widely accepted as a non‐invasive method for quantifying LIC. ³ , ⁴ , ⁵ Because the presence of iron in the liver changes its magnetic susceptibility, SQUID‐BLS is capable of quantifying liver magnetic susceptibility to determine LIC. However, the extremely limited availability of SQUID‐BLS devices (only a few active systems worldwide) is a major obstacle for the widespread use of this technique. Nevertheless, the existing SQUID‐BLS devices represent a useful noninvasive reference for quantification of liver iron overload.

MRI is widely available and has high sensitivity to the abnormal deposition of iron in tissue. Current MRI‐based methods for measuring LIC are based on R₂ or ${\mathrm{R}}_2^{*}$ relaxometry. Multiple studies have already demonstrated strong correlation of R₂ and ${\mathrm{R}}_2^{*}$ with LIC. ⁶ , ⁷ , ⁸ However, these relaxation parameters have an inherently indirect relation to LIC, and therefore, may be confounded by factors other than iron concentration. ⁹ , ¹⁰

In contrast to these indirect relaxometry based methods, MRI‐based QSM ¹¹ , ¹² has the potential to enable direct evaluation of LIC. Although QSM methods have been developed primarily for brain imaging applications, ¹³ , ¹⁴ , ¹⁵ QSM of the abdomen has also been proposed in recent years. ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ QSM‐BLS in patients with liver iron overload has been validated in a limited study, demonstrating strong correlation with SQUID‐BLS. ²⁰ However, previous methods for QSM‐BLS presents substantial technical challenges, including the presence of motion artifacts and a complex susceptibility environment, as well as noise propagation in patients with high iron overload.

More recently, an optimized liver QSM‐BLS reconstruction technique was proposed to address the technical challenges of non‐respiration motion, artifacts, and high LIC. ²¹ Although preliminary results demonstrated a good agreement with R₂‐based LIC quantification, the performance of this method compared to SQUID‐BLS measurements remains unknown. Furthermore, determination of the performance of QSM‐BLS across different acquisition parameters is an important unmet need to demonstrate the reproducibility across various clinical settings.

Therefore, the purpose of this work is to evaluate QSM‐BLS in patients with liver iron overload, acquired with various acquisition protocols at both 1.5 T and 3.0 T, and using SQUID‐BLS as the reference.

2. METHODS

2.1. Study design

In this institutional review board (IRB)‐approved and Health Insurance Portability and Accountability Act (HIPAA)‐compliant prospective study, subjects with known or suspected iron overload were recruited after obtaining informed consent. All subject visits occurred between 2015 and 2019. Each subject's visit included MRI acquisitions for QSM‐BLS estimation at both 1.5 T and 3.0 T, performed approximately within 1 h. In addition, all participants underwent SQUID‐BLS measurements during the study visit or during a contemporaneous clinical visit. The study pipeline is summarized in Figure 1.

FIGURE 1.

Study pipeline overview, including data acquisition module and data post‐processing module

2.2. MRI acquisition and reconstruction

All subjects were imaged using a clinical 1.5 T MRI system (MR450w, GE Healthcare), and a clinical 3.0 T MRI system (MR750, GE Healthcare). Acquisitions were performed with a phased array torso coil or cardiac coil. For QSM‐BLS, breath‐hold multi‐echo 3D spoiled gradient‐echo (SGRE) data were collected for each subject. Because different MRI systems and patients may require different acquisition protocols, multiple QSM acquisitions were completed to evaluate the reproducibility of QSM‐BLS performance. Specifically, eight multi‐echo SGRE imaging protocols with different parameters (flip angle, slice thickness, echo spacing, etc.) were tested at each field strength. The choice of protocols was designed to cover a broad range of reasonable parameters to evaluate tradeoffs in the acquisition of multi‐echo SGRE data within a breath‐hold. For instance, various flip angle choices may lead to different SNR performance of QSM‐BLS. Low flip angle acquisitions are desirable to enable simultaneous fat quantification without T₁ bias, but also may lead to low SNR compared to the Ernst angle. Similarly, spatial resolution effects are important in QSM, because lower resolution may lead to underestimation of susceptibility, ²⁰ , ²² , ²³ whereas higher resolution leads to lower SNR. Importantly, the eight protocols are approximately matched between 1.5 T and 3.0 T (essentially varying the echo spacing's, which are typically shorter at high field strength). Detailed MRI acquisition parameters for each protocol are listed in Table 1.

TABLE 1.

Detailed acquisition parameters for different imaging protocols at 1.5 T and 3.0 T

1.5 T MRI
Pulse Sequence	3D multi‐echo spoiled gradient echo (Axial)
Protocol	1	2	3	4	5	6	7	8
Flip Angle ($°$)	12	5	12	12	5	5	5	5
Slice Thickness (mm)	8	8	10	8	10	8	6	8
No. Echoes	12	12	6	6	6	12	6	6
TE1 (ms)	0.8	0.8	1.2	1.2	1.2	1.2	1.2	1.2
$∆$ TE (ms)	0.8	0.8	2.0	2.0	2.0	2.0	2.0	2.0
TR (ms)	9.8	9.8	13.8	13.8	13.8	12.8	13.8	13.8
FOV (cm2)	40 × 32	40 × 32	40 × 36	40 × 36	40 × 36	40 × 32	40 × 36	40 × 36
Bandwidth (kHz)	±125	±125	±125	±125	±125	±100	±125	±125
Nx × Ny	160 × 128	160 × 128	240 × 160	240 × 160	240 × 160	240 × 128	240 × 160	240 × 160
No. Slices	32	32	28	28	28	24	36	32
Parallel Imaging	2 × 2
Acquisition Time (s)	14	14	18	18	18	17	16	16

3.0 T MRI
Pulse Sequence	3D multi‐echo spoiled gradient echo (Axial)
Protocol	1	2	3	4	5	6	7	8
Flip Angle ($°$)	9	3	9	9	3	3	3	3
Slice Thickness (mm)	8	8	10	8	10	8	6	8
No. Echoes	8	8	6	6	6	8	6	6
TE1 (ms)	0.7	0.7	1.2	1.2	1.2	1.2	1.2	1.2
$∆$ TE (ms)	0.7	0.7	1.0	1.0	1.0	1.0	1.0	1.0
TR (ms)	6.0	6.0	7.7	7.7	8.0	9.2	7.3	7.7
FOV (cm2)	40 × 32	40 × 32	40 × 32	40 × 32	40 × 32	40 × 32	40 × 32	40 × 32
Bandwidth (kHz)	±143	±143	±125	±125	±125	±125	±125	±125
Nx × Ny	144 × 128	144 × 128	256 × 144	256 × 144	256 × 144	240 × 128	256 × 144	256 × 144
No. Slices	32	32	28	32	28	32	36	32
Parallel Imaging	2 × 2
Acquisition Time (s)	12	12	14	14	14	17	16	14

The acquired SGRE multi‐echo images were processed using a complex‐fitting, fat‐corrected algorithm to obtain fat‐only and water‐only images, fat fraction map, B₀ field map $\mathrm{\psi }$, and ${\mathrm{R}}_2^{*}$ map. Subsequently, the B₀ field map and ${\mathrm{R}}_2^{*}$ map were used as input to a recently proposed liver QSM reconstruction algorithm ²¹ based on data‐adaptive regularization to mitigate the effects of motion, high LIC, and other artifacts to generate a magnetic susceptibility map. In this recently developed method, susceptibility $\widehat{\chi }$ was obtained as a solution of regularized dipole inversion problem:

$$\widehat{\chi }=\underset{\chi }{\arg \min }\left({\Vert \mathit{WL}(\psi -D*\chi )\Vert }_2^2+\lambda {\Vert \mathit{P}\chi \Vert }_2^2+\mu {\Vert \mathit{M}\chi \Vert }_2^2\right),$$ (1)

where D is the dipole kernel, L is the Laplace operator, W is a weighting matrix reflecting the reliability of the field map estimate at each voxel. W was selected as a ratio of square root of the sum of squares of signal amplitude across all echoes and residual of the fit of the fat–water model to the multi‐echo signal data. Such choice of W allowed compensating for non‐uniform noise variance in the field map estimates and accounting for field map estimation uncertainties and counteracting amplification and propagation of these errors in the iterative susceptibility estimation. The data consistency term (first summand) was offset by two regularization terms. The first exploits piecewise smoothness of susceptibility maps with regularization operator P based on image gradient as in a previous work. ¹⁹ The second regularization term is motivated by the fact that solution of dipole inversion can be determined only up to an additive constant, therefore, susceptibility quantification is usually performed relative to some reference tissue, such as adipose tissue in liver imaging. Incorporating constraints on the susceptibility of adipose tissue directly into the dipole inversion avoids the need for reference ROI measurements. In this work, the fat mask M was generated by using both the fat fraction map and ${\mathrm{R}}_2^{*}$ map with the following constraints: fat fraction >0.9 and ${\mathrm{R}}_2^{*}$ <300 s⁻¹ for 3.0 T/150 s⁻¹ for 1.5 T (the ${\mathrm{R}}_2^{*}$ constraint is included to avoid areas of low signal or unstable estimation). Regularization parameters $\lambda$ and $\mu$ were optimized empirically individually for each protocol, and then fixed in all experiments. This choice was based on image sharpness, artifact reduction, and algorithm convergence. The resulting regularization parameters were listed here: protocol 1–4 and 6: $\mathrm{\lambda }=5\times {10}^{-5},\mathrm{\mu }=0.1$; protocol 5, 7, and 8: $\mathrm{\lambda }=1\times {10}^{-4},\mathrm{\mu }=0.1$. Details on this formulation and the corresponding optimization algorithm are described in a previous work. ²¹

Additionally, to directly compare QSM‐BLS versus SQUID‐BLS between the new data‐adaptive liver QSM‐BLS method ²¹ and a prior algorithm proposed in Sharma et al., ²⁰ protocol eight acquisition datasets at 3.0 T (most similar acquisition compared to Sharma et al. ²⁰ ) were also reconstructed using the previous QSM‐BLS algorithm. ²⁰ All reconstructions were implemented using MATLAB (The MathWorks).

2.3. SQUID‐BLS acquisition

A dual‐channel SQUID‐BLS system (Ferritometer, Model 5700, Tristan Technologies) was used to provide reference susceptibility measurements. All subjects underwent a previously developed and validated standard measurement procedure with the SQUID‐BLS system within 1 month of the MRI experiments. This standard procedure has been described in detail elsewhere ²⁴ , ²⁵ , ²⁶ and is summarized next. Briefly, each subject underwent an ultrasound examination to determine the best measurement location. Geometric parameters (skin‐to‐liver distance and anterior liver geometry) needed for liver susceptibility quantification from the measured SQUID signals was determined from ultrasound. Patients were placed on a table below the SQUID magnetic field (maximum field 35 mT) and detection coils. SQUID voltages, distance, and position were acquired during a 10‐s breath‐hold vertical scan (15 cm travel). Signals were acquired while the table moved downward at controlled speed, to generate a magnetic flux within the SQUID. These signals were processed using an analytical model with the assumption of ellipsoidal liver and cylindrical thorax geometries. Three measurements were taken using a using a fillable water bag as a reference. The entire measurement took about 0.5 h to complete. This analysis provided an estimate of the magnetic susceptibility of liver relative to water. Subsequently, a report including the liver magnetic susceptibility value of each subject was generated and collected for further analysis. This procedure is similar to that reported in a previous QSM‐SQUID correlation study ²⁰ and is described with further details in the literatures. ²⁴ , ²⁵ , ²⁶

2.4. Measurements and statistical analysis

For QSM‐BLS, a circular region of interest (ROI) with area of ˜15 cm² was placed in the right liver lobe (in Couinaud segment VI or VII) for each acquisition, while avoiding large blood vessels and bile ducts, to measure the liver susceptibility. The rational to choose the ROI in this region is to match the location of SQUID‐BLS sensitivity with better image quality. We also picked two other ROIs to conduct analysis with protocol 8 data, one large circular ROI with area of ˜30 cm², which was centered at the same location as the ROI described above, and another circular ROI with area of ˜15 cm² placed in the left lobe to investigate the effect of ROI selection on liver QSM‐BLS measurements. The locations of these ROIs are illustrated in Figure S1. Note that no reference fat susceptibility measurement was needed in this study as zero‐referencing is embedded into the susceptibility estimation (Equation [1]). Liver ${\mathrm{R}}_2^{*}$ was measured in the same ROI from the ${\mathrm{R}}_2^{*}$ map. All measurements were performed using OsiriX Dicom Viewer (Pixmeo SARL).

Linear mixed‐effects model analysis ²⁷ was performed to evaluate the contributions of specific imaging parameters to QSM‐BLS measurements using the following model: QSM‐BLS $˜$ SQUID‐BLS + field strength + flip angle + slice thickness + echo spacing, in which SQUID‐BLS and different imaging parameters (field strength, flip angle, slice thickness, and echo spacing) are considered as fixed effects. Linear regression analysis was performed between QSM‐BLS and SQUID‐BLS/${\mathrm{R}}_2^{*}$ across the eight different imaging protocols. The observed linear regression parameters were compared to previously published correlation relationships at both 1.5 T and 3.0 T. To compare QSM‐BLS values across field strengths (between 1.5 T and 3.0 T), linear regression and Bland–Altman analysis were applied for each of the eight different imaging protocols. Reproducibility coefficient (RC) between 1.5 T and 3.0 T QSM‐BLS measurements for each protocol was estimated as: $\mathit{RC}=2.77\times \sqrt{\sum _{i=1}^N\frac{{\mathit{SD}}_i^2}{N}}$, where SD is within‐subject standard deviation and N is number of subjects. RC is defined as the smallest significant difference between two repeated measurements taken under different conditions (different field strength in this case). ²⁸ , ²⁹ , ³⁰ All statistical analyses were performed using Python (NumPy, Pandas, Seaborn, and Statsmodels) and MATLAB (The MathWorks).

3. RESULTS

Twenty patients were successfully recruited including 7 children (5 male/2 female; age, 14.3 ± 3.6; BMI, 22.3 ± 3.9 kg/m²) and 13 adults (10 male/3 female; age, 40.5 ± 18.7 years; BMI, 23.9 ± 4.4 kg/m²). For these subjects, the etiologies of iron overload were as follows: sickle cell anemia (n = 1), beta thalassemia (n = 4), Diamond‐Blackfan anemia (n = 5), hemochromatosis (n = 8), sideroblastic anemia (n = 1), and therapy‐related iron overload in high‐grade undifferentiated sarcoma (n = 1). In the subsequent data post‐processing step, one subject's data processing failed in reconstruction because of extreme iron deposition (SQUID‐BLS = 4.2 ppm, which resulted in limited echo signals) and therefore, was excluded from statistical analysis.

Figure 2 shows susceptibility maps from eight different acquisitions obtained at both 1.5 T and 3.0 T for one subject with normal LIC (SQUID‐BLS = 0.67 ppm). Similarly, Figure 3 shows susceptibility maps for one subject with high LIC (SQUID‐BLS = 2.2 ppm). As illustrated in these figures, the new data‐adaptive liver QSM method provides high quality susceptibility maps with different imaging parameters, across various LIC levels.

FIGURE 2.

QSM susceptibility maps at 1.5 T (top) and 3.0 T (bottom) across eight different imaging protocols from a subject with low LIC (SQUID‐BLS = 0.67 ppm). LIC, liver iron concentration; SQUID, superconducting quantum interference device; BLS, biomagnetic liver susceptometry

FIGURE 3.

QSM susceptibility maps at 1.5 T (top) and 3.0 T (bottom) across eight different imaging protocols from a subject with high LIC (SQUID‐BLS = 2.2 ppm). LIC, liver iron concentration; SQUID, superconducting quantum interference device; BLS, biomagnetic liver susceptometry

Linear mixed‐effects modeling results evaluating the relation between QSM‐BLS and SQUID‐BLS, as well as different MR imaging parameters (field strength, flip angle, slice thickness, and echo spacing) are detailed in Table 2. SQUID‐BLS has a highly significant effect (p < 0.01) to slope (coefficient $=$ 0.584). Field strength has a significant effect (p < 0.01), but with moderate contribution to slope (coefficient = 0.020). The remaining MR imaging parameters have no significant effect on QSM‐BLS measurements (p > 0.01).

TABLE 2.

Linear mixed‐effects model analysis of QSM‐BLS in terms of SQUID‐BLS and different MR imaging parameters

	Intercept (ppm)	SQUID‐BLS (ppm)	Field strength (T)	Flip Angle ($°$)	Slice Thickness (mm)	Echo Spacing (ms)
Coefficient	$-$0.583	0.584	0.020	0.002	$-$0.002	$-$0.002
Standard error	0.119	0.064	0.004	0.001	0.003	0.001
P	<0.01	<0.01	<0.01	0.025	0.441	0.091

Abbreviations: BLS, biomagnetic liver susceptometry; SQUID, superconducting quantum interference device

Figure 4 demonstrates strong correlation between 1.5 T QSM‐BLS versus SQUID‐BLS (r ² range, [0.82, 0.84]) and 3.0 T QSM‐BLS versus SQUID‐BLS (r ² range, [0.77, 0.85]). The linear regression relationship results (slope, intercept, and r ²) across different protocols are reported in Table 3.

FIGURE 4.

High linear correlation relationships observed between QSM‐BLS and SQUID‐BLS at 1.5 T (slope range, [0.54, 0.59]; intercept range, [$-$0.56, $-$0.52] ppm; r ² range, [0.82, 0.84]) and 3.0 T (slope range, [0.59, 0.62]; intercept range, [$-$0.60, $-$0.55] ppm; r ² range, [0.77, 0.85]) across eight different imaging protocols. A previous study by Sharma et al.²⁰ showed strong correlation between 3.0 T QSM‐BLS and SQUID‐BLS with linear regression parameters (slope = 0.49, intercept = $-$0.22 ppm, r ² = 0.88). BLS, biomagnetic liver susceptometry; SQUID, superconducting quantum interference device

TABLE 3.

Linear regression results (slope, intercept, and r ²) comparing QSM‐BLS and SQUID‐BLS across different protocols at 1.5 T and 3.0 T

Field Strength	Protocol	Slope	Intercept (ppm)	r 2
1.5 T	1	0.57 $\pm$ 0.07	$-$0.54 $\pm$ 0.12	0.82
	2	0.55 $\pm$ 0.06	$-$0.54 $\pm$ 0.10	0.84
	3	0.57 $\pm$ 0.07	$-$0.54 $\pm$ 0.12	0.82
	4	0.59 $\pm$ 0.06	$-$0.56 $\pm$ 0.12	0.84
	5	0.58 $\pm$ 0.06	$-$0.55 $\pm$ 0.12	0.83
	6	0.54 $\pm$ 0.06	$-$0.52 $\pm$ 0.11	0.82
	7	0.58 $\pm$ 0.06	$-$0.55 $\pm$ 0.12	0.82
	8	0.56 $\pm$ 0.06	$-$0.54 $\pm$ 0.11	0.84
3.0 T	1	0.61 $\pm$ 0.08	$-$0.57 $\pm$ 0.14	0.80
	2	0.61 $\pm$ 0.07	$-$0.59 $\pm$ 0.13	0.84
	3	0.59 $\pm$ 0.08	$-$0.55 $\pm$ 0.15	0.77
	4	0.62 $\pm$ 0.07	$-$0.60 $\pm$ 0.13	0.83
	5	0.59 $\pm$ 0.07	$-$0.56 $\pm$ 0.13	0.82
	6	0.60 $\pm$ 0.06	$-$0.57 $\pm$ 0.12	0.85
	7	0.61 $\pm$ 0.07	$-$0.59 $\pm$ 0.12	0.84
	8	0.61 $\pm$ 0.08	$-$0.57 $\pm$ 0.14	0.80

Abbreviations: BLS, biomagnetic liver susceptometry; SQUID, superconducting quantum interference device.

For one of the protocols acquired in this work (protocol 8 at 3.0 T), a strong correlation (slope = 0.69, intercept = $-$0.71 ppm, r ² = 0.87) between 3.0 T QSM‐BLS obtained with a previous liver QSM reconstruction method ²⁰ and SQUID‐BLS was observed as well.

As shown in Figure 5, strong correlation was observed between 1.5 T QSM‐BLS versus ${\mathrm{R}}_2^{*}$ (r ² range, [0.94, 0.99]) and 3.0 T QSM‐BLS versus ${\mathrm{R}}_2^{*}$ (r ² range, [0.93, 0.99]). The linear regression relationship results (slope, intercept, and r ²) across different protocols are reported in Table S1.

FIGURE 5.

High linear correlation relationships observed between QSM‐BLS and ${\mathrm{R}}_2^{*}$ at 1.5 T (slope range, [0.0044, 0.0050] ppm/s⁻¹; intercept range, [$-$0.52, $-$0.45] ppm; r ² range, [0.94, 0.99]); and 3.0 T (slope range, [0.0022, 0.0026] ppm/s⁻¹; intercept range, [$-$0.49, $-$0.43] ppm; r ² range, [0.93, 0.99]) across eight different imaging protocols. Close agreement of both slope and intercept was observed with a previously reported relationship (1.5 T: slope = 0.0055 ppm/s⁻¹, intercept = $-$0.51 ppm, r ² = 0.94; 3.0 T: slope = 0.0028 ppm/s⁻¹, intercept intercept = $-$0.54 ppm, r ² = 0.93) from Sharma et al.²⁰ The slope agreed with another previously reported relationship at 3.0 T (slope = 0.0030 ppm/s⁻¹, intercept = 0.07 ppm, r ² = 0.80) from Li et al.³¹ BLS, biomagnetic liver susceptometry

Strong correlation (r ² range = [0.98, 0.99]) was observed between susceptibility estimates from QSM‐BLS at 1.5 T and 3.0 T in Figure 6. Table 4 provides the linear regression relationship results (slope, intercept, and r²) across different protocols. Bland–Altman analysis comparing QSM‐BLS at 1.5 T and 3.0 T indicates low bias within [$-$0.17, 0.36] ppm. High reproducibility between 1.5 T and 3.0 T QSM‐BLS were observed for each of the eight different protocols and across protocols (protocol 1 RC, 0.15 ppm; protocol 2 RC, 0.17 ppm; protocol 3 RC, 0.15 ppm; protocol 4 RC, 0.14 ppm; protocol 5 RC, 0.13 ppm; protocol 6 RC, 0.16 ppm; protocol 7 RC, 0.14 ppm; protocol 8 RC, 0.21 ppm).

FIGURE 6.

High linear correlation relationships (slope range, [1.00, 1.11]; intercept range, [0.01, 0.02] ppm; r ² range, [0.98, 0.99]) observed between 3.0 T QSM‐BLS and 1.5 T QSM‐BLS across eight different imaging protocols (left). Bland–Altman analysis of QSM‐BLS measurements at 1.5 T and 3.0 T showed low bias within [$-$0.17, 0.36] ppm (right). BLS, biomagnetic liver susceptometry

TABLE 4.

Linear regression results (slope, intercept, and r ²) comparing 3.0 T QSM‐BLS and 1.5 T QSM‐BLS across different protocols

Protocol	Slope	Intercept (ppm)	r 2
1	1.07 $\pm$ 0.03	0.01 $\pm$ 0.02	0.99
2	1.11 $\pm$ 0.02	0.02 $\pm$ 0.01	0.99
3	1.06 $\pm$ 0.03	0.01 $\pm$ 0.02	0.99
4	1.03 $\pm$ 0.03	0.01 $\pm$ 0.02	0.99
5	1.00 $\pm$ 0.03	0.01 $\pm$ 0.02	0.99
6	1.09 $\pm$ 0.02	0.01 $\pm$ 0.02	0.99
7	1.03 $\pm$ 0.03	0.01 $\pm$ 0.02	0.99
8	1.10 $\pm$ 0.03	0.02 $\pm$ 0.03	0.98

Abbreviation: BLS, biomagnetic liver susceptometry.

4. DISCUSSION

This work successfully demonstrated the feasibility of liver QSM‐BLS across iron levels and MRI acquisition parameters at both 1.5 T and 3.0 T. Standard SQUID‐BLS measurements were obtained as the reference to validate QSM‐BLS. High correlation was observed between QSM‐BLS at both field strengths, and SQUID‐BLS. This work confirms and extends previous works on different MR vendors and with multiple field strengths. On further optimization and validation, these results may have implications for the reliable and reproducible quantification of liver iron overload.

High linear correlation between 3.0 T QSM‐BLS and SQUID‐BLS was observed (slope range, [0.59, 0.61]; intercept range, [$-$0.60, $-$0.55] ppm; and r ² range, [0.77, 0.85]), which showed a small discrepancy with previous reported relationship ²⁰ (slope = 0.49, intercept = $-$0.22 ppm, r ² = 0.88). Note that 3.0 T protocol 8 dataset in this work was reconstructed using same algorithm proposed in Sharma et al., ²⁰ which resulted in a similar underestimated relationship (slope = 0.69, intercept = $-$0.71 ppm, r ² = 0.87). Therefore, the difference in slope between the current study and a previous study, ²⁰ may be caused in part by the different QSM reconstruction algorithm. Additionally, the reconstruction method used in this work had fat tissue zero‐referencing embedded into the regularization, thereby avoiding the need for “boundary” measurements (where a reference ROI is placed in the nearby subcutaneous fat). With the data‐adaptive regularization method used in this work, ²¹ only a single ROI was needed in the liver to measure susceptibility. In Figure S2, ROI analysis results based on different selections showed similar correlation relationship between QSM‐BLS and SQUID‐BLS. Therefore, the cause of the underestimation of QSM‐BLS relative to SQUID‐BLS (slope <1) remains unknown, although is reproducible. In previous works, ²² , ²³ the spatial resolution in QSM‐BLS acquisitions has been shown to be a potential source of bias. It is also possible that the background field removal technique applied in QSM‐BLS method, partial volume effects, selections of regularization parameters, and embedded susceptibility prior eliminates some of the susceptibility information, leading to the underestimation. In addition, residual differences (because of different devices) may exist between the reference SQUID measurements used in this study compared to previous studies. ²⁰ Although SQUID‐BLS and QSM‐BLS both aim to measure liver susceptibility, SQUID‐BLS provides a single global liver susceptibility measurement, without providing information on the spatial distribution of susceptibility. This may introduce variability in SQUID‐BLS measurements because of liver heterogeneity and may explain part of the apparent discrepancies. Further studies may be required to fully characterize these discrepancies.

High linear correlation was observed between QSM‐BLS and ${\mathrm{R}}_2^{*}$ at 1.5 T and 3.0 T. At 1.5 T, this correlation has good agreement with previous work ¹⁹ (slope = 0.0055 ppm/s⁻¹, intercept = $-$0.51 ppm, r ² = 0.94). At 3.0 T, this study also has good agreement with Sharma et al. ¹⁹ (slope = 0.0028 ppm/s⁻¹, intercept = $-$0.54 ppm, r ² = 0.93), but there is discrepancy in the intercept compared to Li et al. ³¹ (slope = 0.0030 ppm/s⁻¹, intercept = 0.07 ppm, r ² = 0.80) at low iron level. This intercept may depend on the QSM‐BLS regularization strategy as well as the reference tissue chosen for susceptibility measurements (no reference in this work vs. latissimus dorsi muscle as reference in the work by Li et al.) ³¹ Importantly, this work confirmed a close correlation between QSM‐derived liver susceptibility measurements and liver ${\mathrm{R}}_2^{*}$. It is possible that the relationship between susceptibility and ${\mathrm{R}}_2^{*}$ may depend on the etiology of iron overload, as well as on the chelation status. ⁷ , ⁸ , ¹⁰ However, this study was not powered to evaluate such differences across different types of patients. If this is demonstrated in future studies, then QSM and ${\mathrm{R}}_2^{*}$ mapping, which are obtained from the same acquisition, may provide complementary information regarding iron deposition.

In this work, the reproducibility of liver QSM‐BLS was studied between field strengths. High reproducibility (reproducibility coefficients, 0.13–0.21 ppm) was observed between 1.5 T QSM‐BLS and 3.0 T QSM‐BLS across different protocols. Note the slightly higher susceptibility observed in 3.0 T may be driven by high iron cases because of rapid signal decay in the presence of high ${\mathrm{R}}_2^{*}$ at 3.0 T. Based on results from linear mixed‐effects modeling, field strength has a significant effect to QSM‐BLS. However, the difference in slope between QSM‐BLS at each field strength and SQUID‐BLS is small (slope at 1.5 T is 0.02 lower than at 3.0 T).

This work has several limitations. Although the validation study included eight different axial acquisitions, these protocols do not cover all possible combinations of imaging parameters. Additional imaging protocols (e.g., coronal/sagittal acquisition) may be interesting to examine in future work. Nevertheless, promising results were demonstrated for the selected protocols used in this work. This study only used Cartesian acquisitions, which limit the shortest TE achievable to ˜0.7 ms. Liver QSM in the presence of high iron with these acquisitions is challenging because of the very rapid ${\mathrm{R}}_2^{*}$ signal decay and led to one subject with high LIC being discarded from data analysis in this study. Non‐Cartesian acquisition techniques (e.g., radial or spiral), ³² , ³³ which enable much shorter echo time, may be promising as part of future work.

In summary, we have demonstrated reliable and reproducible liver QSM‐BLS across liver iron levels and various MRI acquisition protocols at both 1.5 T and 3.0 T using SQUID‐BLS as the reference.

FUNDING INFORMATION

National Institutes of Health, Grant/Award Numbers: R01‐DK117354, R01‐DK100651, and R01‐DK088925

Supporting information

Table S1. Linear regression results (slope, intercept, and r ²) comparing QSM‐BLS and ${\mathrm{R}}_2^{*}$ across different protocols at 1.5 T and 3.0 T.

Figure S1. Three ROIs shown in one subject QSM susceptibility map.

Figure S2. High linear correlation relationships observed between QSM‐BLS and SQUID‐BLS at 1.5T (slope range: [0.55, 0.57], intercept range: [$-$0.56, $-$0.54] ppm, r ² range: [0.84, 0.85]) and 3.0 T (slope range: [0.61, 0.63], intercept range: [$-$0.60, $-$0.57] ppm, r ² range: [0.80, 0.84]) across three different ROIs.

Zhao R, Velikina J, Reeder SB, Vasanawala S, Jeng M, Hernando D. Validation of liver quantitative susceptibility mapping across imaging parameters at 1.5 T and 3.0 T using SQUID susceptometry as reference. Magn Reson Med. 2023;89:1418‐1428. doi: 10.1002/mrm.29529