Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Mar 1.
Published in final edited form as: Magn Reson Med. 2021 Jan 11;85(5):2608–2621. doi: 10.1002/mrm.28610

Free-Breathing R2* Mapping of Hepatic Iron Overload in Children Using 3D Multi-Echo UTE Cones MRI

Youngwook Kee 1, Christopher M Sandino 1, Ali B Syed 1, Joseph Y Cheng 1, Ann Shimakawa 2, Timothy J Colgan 3, Diego Hernando 3, Shreyas S Vasanawala 1
PMCID: PMC8886621  NIHMSID: NIHMS1777036  PMID: 33432613

Abstract

PURPOSE

To enable motion-robust, ungated, free-breathing R2* mapping of hepatic iron overload in children with 3D multi-echo UTE cones MRI.

METHODS

A golden-ratio re-ordered 3D multi-echo UTE cones acquisition was developed with chemical-shift encoding (CSE). Multi-echo complex-valued source images were reconstructed via gridding and coil combination, followed by confounder-corrected R2* (=1/T2*) mapping. A phantom containing 15 different concentrations of gadolinium solution (0-300 mM) was imaged at 3T. 3D multi-echo UTE cones with an initial TE of 0.036 ms and Cartesian CSE-MRI (IDEAL-IQ) sequences were performed. With IRB approval, 85 subjects (81 pediatric patients with iron overload + 4 healthy volunteers) were imaged at 3T using 3D multi-echo UTE cones with free breathing (FB Cones), IDEAL-IQ with breath holding (BH Cartesian) and free breathing (FB Cartesian). Overall image quality of R2* maps was scored by two blinded experts and compared by a Wilcoxon rank sum test. For each pediatric subject, the paired R2* maps were assessed to determine if a corresponding artifact-free 15 mm region-of-interest (ROI) could be identified at a mid-liver level on both images. Agreement between resulting R2* quantification from FB Cones and BH/FB Cartesian was assessed with Bland-Altman and linear correlation analyses.

RESULTS

ROI-based regression analysis showed a linear relationship between gadolinium concentration and R2* in IDEAL-IQ (y=8.83x−52.10, R2=0.995) as well as in cones (y=9.19x−64.16, R2=0.992). ROI-based Bland-Altman analysis showed that the mean difference (MD) was 0.15% and the SD was 5.78%. However, IDEAL-IQ R2* measurements beyond 200 mM substantially deviated from a linear relationship for IDEAL-IQ (y=5.85x+127.61, R2=0.827), as opposed to cones (y=10.87x−166.96, R2=0.984). In vivo, FB Cones R2* had similar image quality with BH and FB Cartesian in 15 and 42 cases, respectively. FB Cones R2* had better image quality scores than BH and FB Cartesian in 3 and 21 cases, respectively, where BH/FB Cartesian exhibited severe ghosting artifacts. ROI-based Bland-Altman analyses were 2.23% (MD) and 6.59% (SD) between FB Cones and BH Cartesian and were 0.21% (MD) and 7.02% (SD) between FB Cones and FB Cartesian, suggesting a good agreement between FB Cones and BH (FB) Cartesian R2*. Strong linear relationships were observed between BH Cartesian and FB Cones (y=1.00x+1.07, R2=0.996) and FB Cartesian and FB Cones (y=0.98x+1.68, R2=0.999).

CONCLUSION

Golden-ratio re-ordered 3D multi-echo UTE Cones MRI enabled motion-robust, ungated, free-breathing R2* mapping of hepatic iron overload, with comparable R2* measurements and image quality to BH Cartesian, and better image quality than FB Cartesian.

Keywords: Hepatic iron overload, free-breathing liver R2* mapping, 3D multi-echo UTE cones k-space sampling trajectory, chemical-shift-encoded MRI, confounder-corrected R2*

INTRODUCTION

Iron overload is caused by excess intestinal absorption (e.g., hereditary hemochromatosis) or repeated blood transfusions (e.g., transfusion hemosiderosis) 1. Excess body iron may lead to organ dysfunction and damage, necessitating treatment to reduce body iron stores 2; 3. Accurate assessment of the body iron is important to guide expensive iron-chelation therapy that can have toxic side-effects 4.

Liver iron concentration, or LIC [mg Fe/g dry tissue weight], has a strong correlation with the total body iron stores 4; 5. Liver biopsy is often performed with subsequent atomic absorption spectrophotometry as the gold standard to measure LIC. However, liver biopsy is invasive, subject to sampling variability, and has risk of uncontrolled bleeding 6. Thus, MRI relaxometry (R2 or R2*) is clinically a noninvasive alternative. R2 and R2* are superior to liver biopsy in LIC accuracy 7, making MRI the standard of care.

Despite clinical use, R2-based LIC quantification techniques are time consuming, requiring scan time of 5-20 minutes. Commercially available chemical-shift-encoded (CSE) MRI techniques for R2* mapping require much shorter scan times with whole liver gradient echo acquisitions – typically ~20 sec with breath holding and 1 to 5 min with free breathing.

However, there are two major challenges in the commercially available CSE-MRI techniques for hepatic R2* mapping. One of the challenges is respiratory motion. Current CSE-MRI techniques use 3D Cartesian k-space data sampling, usually with breathing-holding. Breath-holding is challenging for many children and some adults, impractical for children under anesthesia; poor breath-holding often leads to poor R2* image quality due to ghost artifacts along the phase encoding direction (Figure 1a). Signal averaging (multi-NEX) is often utilized in Cartesian CSE-MRI to enable free-breathing acquisitions; however, R2* images still often exhibit respiratory motion artifacts (Figure 1b). The other inherent challenge in R2* mapping with the current Cartesian CSE-MRI techniques is rapid T2* decay, particularly at 3T, with high degrees of iron overload, often seen in secondary iron overload 8. This yields inadequate SNR of the first echo due to a relatively long initial TE (~1 ms), making R2* measurements unreliable. A representative example in Figure 1 (c) highlights a patient with moderate iron overload showing a large SD of R2* measurements at 3T.

Figure 1.

Figure 1.

Open in a new tab

Challenges in conventional Cartesian CSE-MRI in liver R2* mapping. Top row: Cartesian CSE-MRI is shown in (a) with a single breath-hold and in (b) with signal averaging that enable free-breathing. Bottom row: free-breathing liver R2* mapping of 3D multi-echo UTE cones (FB Cones) was demonstrated to produce R2* maps without ghosting artifacts (d) and (e). R2* maps of a patient with iron overload shown in (c) exhibits noisy and appears to have unreliable R2* measurements within the ROI (red) as it has a large SD. The R2* map in (f) exhibits less noise and appears to produce more reliable R2* measurements within the ROI (red) as it has a smaller SD. Detailed imaging parameters are as follows: (a) 2.4X ACC, NEX=0.75, scan time=0:23, and res=1.17x1.17x6 mm3; (b) 2X ACC, NEX=4, scan time=2:53, and res=1.25x1.25x10 mm3; (c) 4X ACC, NEX=0.75, scan time=0:15, and rex=1.17x1.17x6 mm3; (d) no ACC, NEX=1, scan time=3:39, and res=1.44x1.44x3 mm3; (e) no ACC, NEX=1, scan time=6:30, and res=2.3x2.3x3 mm3; (f) no ACC, NEX=1, scan time=4:16, and rex=1.48x1.48x6 mm3. Note that the subject shown in (c) and (d) additionally underwent 1.5T to receive FerriScan, i.e., FerriScan LIC = 8.4 mg Fe/g dry tissue weight, and was excluded from our study due to different slice thickness of FB Cones.

In this work, a 3D multi-echo UTE cones MRI technique was developed to address these challenges in R2* mapping of hepatic iron overload in children, enabling motion-robust, fully ungated, and free-breathing MRI. The developed technique was validated against a commercially available Cartesian CSE-MRI on a phantom and as well as in human subjects.

METHODS

3D Multi-Echo UTE Cones Trajectory

A spoiled gradient echo (spoiled-GRE) sequence with a 3D cones k-space trajectory 9 was extended to a 3D multi-echo UTE cones acquisition as shown in Figure 2. A spiral-like cone interleaf traverses from the center of k-space along the conical surface and rewinds back to the center of k-space using time-optimal gradient waveforms 10. This readout/rewinding cycle repeats multiple times to acquire multiple echoes in a time-interleaved fashion with multiple excitations. Once this cycle completes, the next cone interleaf is acquired with the same cycle until the Nyquist sampling criterion is met. In a sequential 3D cones acquisition, the cone interleaves are acquired in order from “north” to “south” pole of the sampled k-space sphere. To improve motion robustness, the order of cone interleaves is determined by permuting the sequentially ordered cone interleaves by the golden-ratio 11.

Figure 2.

Figure 2.

Open in a new tab

3D multi-echo UTE Cones k-space sampling trajectory with the golden-ratio re-ordering. In (a) and (b), 15 conical surfaces spanning from the “south” to the “north” pole of the k-space are displayed. These surfaces were subsampled at every 15th conical surface from the south to the north pole out of the total 211 conical surfaces determined by a given FOV and resolution. The number of cone interleaves on these surfaces are determined to meet the Nyquist sampling criterion based on the readout duration, then these cone interleaves are re-ordered using the golden-ratio permutation as displayed in (c). The gradient waveforms shown in (d) are associated with the cone interleaf in red in (a), (b), and (c), where 6 echoes are acquired with two shots and an echo train length of 3 in a time-interleaved fashion. Readout duration of each cone interleaf is ~1 ms.

Image Reconstruction and Confounder-Corrected R2* Mapping

Complex-valued source (magnitude and phase) images were reconstructed from k-space raw data via gridding reconstruction implemented in SigPy, a Python package for GPU-accelerated image reconstruction 12. Coil images were combined using the adaptive matched filter method 13. A confounder-corrected R2* mapping technique was subsequently performed using nonlinear least-squares fitting of the complex-valued source images together with six-peak fat modeling, which produces fat-corrected R2*, B0 field, and proton density fat fraction (PDFF) maps 14 (Figure 3).

Figure 3.

Figure 3.

Open in a new tab

Image reconstruction and R2* mapping pipeline. Gridding reconstruction and coil combination are performed to generate multi-echo complex-valued source images (magnitude and phase) from the k-space raw data with its coordinates (step 1). These source images are subsequently processed via confounder-corrected R2* mapping (nonlinear least squares fitting) in consideration of the presence of fat and B0 field inhomogeneity (step 2).

Cartesian CSE-MRI

For comparison, a commercially available multi-echo spoiled-GRE sequence (IDEAL-IQ, GE Healthcare, Waukesha, WI) was used with the confounder-corrected R2* mapping technique. IDEAL-IQ is a CSE 3D volumetric multi-echo spoiled-GRE imaging sequence designed to acquire multiple images with suitable echo spacing. With Cartesian k-space data sampling, IDEAL-IQ allows parallel imaging, partial Fourier imaging, or signal averaging (multi-NEX imaging). These features provide scan time flexibility for a single breath-hold or a free-breathing scan with signal averaging. IDEAL-IQ runs with its built-in software product for image reconstruction and parameter mapping. After each IDEAL-IQ scan, volumetric R2* and PDFF are produced and are readily available on the scanner in DICOM file format. Note that storing B0 field map and complex-valued source images in DICOM is optionally available. In this work, when complex-valued source images were not available, R2* produced by the product reconstruction was used for subsequent analyses.

Phantom Study

Gadolinium solution (Gadavist, Bayer Pharmaceuticals, Wayne, NJ) was diluted with saline (0.9% Sodium Chloride Injection, USP, Hospira Inc, IncLake Forest, IL) into 15 different concentrations ranging from 0 to 300 mM. Each of the diluted gadolinium solutions (0, 6.25, 12.5, 25, 50, 75, 100, 125, 150, 175, 200, 225, 250, 275, and 300 mM) was contained in 50 mL Falcon tubes and placed in a polystyrene foam container as displayed in Figure 4 (a). Then, the phantom was imaged on a 3T clinical MRI scanner (MR750, GE Healthcare, Waukesha, WI) with 3D multi-echo UTE cones and IDEAL-IQ sequences. For 3D multi-echo UTE cones, a set of 6 echoes was acquired with varying echo times, i.e., 6 shots and an echo train length (ETL) of 1, which achieves an echo spacing of 0.2 ms. For IDEAL-IQ, a set of 6 echoes was acquired with varying echo times, 2 shots and an ETL of 3, which achieves the minimum achievable echo spacing of 0.73 ms. The initial TE of 3D multi-echo UTE cones was 0.036 ms, and the minimum achievable initial TE of IDEAL-IQ was 0.79 ms. The image quality of computed R2* maps was visually compared. For quantitative analysis, circular regions-of-interest (ROIs) with a radius of 10 mm were placed in the center of each of the vials in the R2* maps, and their means and SD were calculated to perform ROI-based linear correlation and Bland-Altman analyses.

Figure 4.

Figure 4.

Open in a new tab

Phantom study. Configuration of a phantom with different concentrations of gadolinium solutions ranging from 0 to 300 mM is shown in (a). R2* maps (resolution = 2x2x2 mm3) of 3D UTE cones and Cartesian (IDEAL-IQ) are shown in (b) and (c), respectively. The R2* maps show a similar visual appearance between 0 to 200 mM of [Gd] indicated with the yellow dotted bounding boxes. However, beyond 200 mM, Cartesian R2* image quality appears to be noisy and the apparently linear relationship between R2* and [Gd] breaks down.

In Vivo Study – Patients and Healthy Volunteers

With IRB approval and informed consent/assent, 85 subjects (81 pediatric patients with known and suspected hepatic iron overload (9.6 +/− 6.4 years old, 45M and 36F) and 4 healthy adult volunteers (34.2 +/− 6.1 years old, 4M)) were imaged on three 3T clinical MRI scanners (two MR750 and one PET/MR, GE Healthcare, Waukesha, WI) with 3D multi-echo UTE cones and Cartesian CSE-MRI (IDEAL-IQ) sequences. 3D multi-echo UTE cones MRI was performed with free breathing (FB Cones), and Cartesian CSE-MRI was performed in a clinical routine fashion, with either breath-holding or free breathing based on technologists’ decision and subjects’ ability to hold their breath. Cartesian CSE-MRI with breath-holding utilized parallel imaging along the phase and slice encoding directions (R=2x2, i.e., 4X ACC) and partial Fourier imaging (= 0.75) to cover the entire liver volume within a single breath-hold (BH Cartesian). Cartesian CSE-MRI with free breathing was performed with signal averaging (NEX = 2 to 6) and 1 to 4X acceleration (FB Cartesian). Detailed imaging parameters for in vivo studies are listed in Table 1. Note that low flip angles (FAs) were used as this enables joint quantification of R2* and PDFF, minimizing T1-related bias in fat quantification 22. 22 subjects (18 patients + 4 volunteers) out of 85 underwent FB Cones and BH Cartesian, and 63 subjects underwent FB Cones and FB Cartesian. One of the patients who underwent FB Cones and BH Cartesian was also imaged on a 1.5T clinical scanner (Artist, GE Healthcare, Waukesha, WI) for an FDA-approved R2-based LIC quantification method, FerriScan LIC (Resonance Health Ltd), as shown in Figure 1 (c).

Table 1.

Imaging parameters for in vivo study

FB Cones BH Cartesian FB Cartesian
Plane/Mode/Pulse Ax/3D/SPGR Ax/3D/SPGR (IDEAL-IQ) Ax/3D/SPGR (IDEAL-IQ)
Readout cones R/L (unipolar) R/L (unipolar)
#TEs 6 (#shots=2; ETL=3) 6 (#shots=2; ETL=3) 6 (#shots=2; ETL=3)
TE1 / TE6 [ms] 0.036 / 5.096 1.132 / 5.63 – 6.63 1.132 / 5.63 – 6.63
ΔTE [ms] 1.012 0.9 – 1.1 0.9 – 1.1
TR [ms] 10.1 – 11.4 7.3 – 8.5 6.2 – 13.9
FA [deg] 3 3 3 – 5
NEX 1 0.75 (partial Fourier) 2 – 6 (multi-NEX)
rBW [Hz/Px] 1250 – 1389 977 868
In-plane res [mm3] 1.2×1.2 – 2.67×2.67 1.17×1.17 – 1.6×1.6 0.8×0.8 – 1.9×1.9
Slice thickness [mm] 3 6 6 – 10
Number of slices 32 – 90 16 – 48 12 – 40
Acceleration No ACC 4X ACC No/2X ACC
Coil 32-ch cardiac/torso/body 32-ch cardiac/torso/body 32-ch cardiac/torso/body
Acquisition time 2:54 – 6:30 0:17 – 0:23 1:08 – 5:55
Open in a new tab

Note that as opposed to BH Cartesian, complex-valued source images were not available in our FB Cartesian protocol. Therefore, online IDEAL-IQ software product reconstruction was used for R2* mapping.

Reader Study – Overall Image Quality Assessment

Image quality of the total 162 volumetric R2* maps (81 patients with 81 FB Cones R2* + either 63 FB Cartesian or 18 BH Cartesian R2* maps) was assessed by 2 board certified radiologists with five and eighteen years of experience in body MR interpretation. The raters were blinded to image type (FB Cones, FB Cartesian, and BH Cartesian), and the total 162 volumetric R2* maps were presented in randomized order to each of the raters independently. The raters were able to scroll through all axial slices for each dataset. Image quality was scored on a 5-point scale: 1 = no region of liver can be evaluated; 2 = <1/3 of liver can be evaluated; 3 = 1/3 to 2/3 of liver can be evaluated; 4 = >2/3 of liver can be evaluated; 5 = all of liver can be evaluated. Median scores of FB Cones and BH/FB Cartesian were computed for each rater, along with their interquartile ranges. A Wilcoxon signed-rank test was performed to compare FB Cones scores to BH/FB Cartesian scores for each rater. Two-way mixed, single measures, consistency intraclass correlation coefficient (ICC) was calculated with its 95% confidence interval (CI) to assess inter-rater agreement.

In Vivo Study – Comparison between FB Cones and BH/FB Cartesian R2* Quantification within ROIs

For quantitative comparison of ROI-based R2* measurements, the total 81 datasets (63 datasets with FB Cones and FB Cartesian + 18 datasets with FB Cones and BH Cartesian) were first visually assessed independently from the reader study. For each of the 81 datasets, two sets of images, i.e., one from FB Cones and the other from BH/FB Cartesian, were assessed side-by-side to determine whether an artifact-free region can be identified for ROI placement on the corresponding axial mid-liver slices of FB Cones and BH/FB Cartesian. Datasets for which this ROI placement was not possible were excluded from further analysis. For the remaining datasets, circular ROIs with a radius of 15 mm were placed on these mid-liver slices, avoiding artifacts and large vessels, such as the portal vein, to compute mean and SD of R2* within the ROIs. The resulting measurements were then evaluated with Bland-Altman analysis and scatter plots with linear correlation analysis.

RESULTS

Acquisition Times

FB Cones took 2:54 to 6:30 with a median acquisition time of 4:23 with the interquartile range (IQR) of 1:14. BH Cartesian took 0:17 to 0:23 (median 0:18, IQR 0:04). FB Cartesian took 1:08 to 5:55 (median 2:24, IQR 2:29). These are listed in Table 1.

Computation Time

Gridding reconstruction in SigPy took 134.7 +/− 29.8 sec on a Titan RTX graphics card (NVIDIA, Santa Clara, CA) for 6 echoes and 32-channel 3D cones k-space data. Confounder-corrected R2* mapping from the reconstructed complex-valued coil images took 251.3 +/− 25.9 sec on MATLAB R2018b (MathWorks, Natick, MA) with an Intel Core i7-7567U CPU. For Cartesian CSE-MRI, R2* and coil-combined complex-valued source images were available from product GE reconstruction within about 1 minute.

Phantom Study

R2* maps from 3D multi-echo UTE cones and IDEAL-IQ had similar visual appearance between 0 to 200 mM of gadolinium concentration, [Gd], as shown in Figure 4 (b) and (c) with yellow dotted bounding boxes. For the corresponding gadolinium vials (0, 6.25, 12.5, 25, 50, 75, 100, 125, 150, 175, and 200 mM), mean values (and SD) of the ROI-based R2* measurements for 3D multi-echo UTE cones were 9 (5.9), 45 (10.4), 72 (11.2), 169 (9.3), 370 (8.6), 543 (15.2), 783 (31.1), 1028 (43.7), 1284 (48.9), 1572 (41.9), and 1933 (72.8) s−1. Mean values (and SD) of the ROI-based R2* measurements for IDEAL-IQ were 3 (2.6), 50 (11.1), 70 (4.4), 159 (5.7), 350 (6.4), 552 (16.4), 784 (25.2), 1016 (45.0), 1271 (67.8), 1504 (142.2), and 1785 (227.0) s−1. ROI-based regression analyses showed a linear relationship between gadolinium concentration and R2* in 3D multi-echo UTE cones (y = 9.19x – 64.16, R2 = 0.992) as well as in IDEAL-IQ (y = 8.83x – 52.10, R2 = 0.995) as shown in Figure 5 (a). As shown in Figure 5 (b), Bland-Altman plot, the bias (mean difference, MD) was 0.15% and the SD was 5.78%. The 95% confidence interval (CI) of agreement limits (limits of agreement, LoA) was [−11.2%, 11.5%] across the range of measured R2* values (6.25 to 200 mM of [Gd]), within which nine data points except for one point with 6.25 mM of [Gd] fell, suggesting a good agreement between 3D multi-echo UTE cones and IDEAL-IQ in ROI-based R2* measurements. A strong correlation between these methods was also observed via the linear correlation plot in Figure 5 (c), where y = 1.04x – 13.22, R2 = 0.999.

Figure 5.

Figure 5.

Open in a new tab

In vitro ROI-based linear regression and Bland-Altman plots. ROI-based linear regression plots are shown in (a), showing a linear relationship between [Gd] ranging from 0 to 200 mM and R2*. Bland-Altman and linear regression analyses shown in (b) and (c) suggest a good agreement between the two methods as the bias was 0.15%, the SD was 5.78%, i.e., LoA = [−11.2%, 11.5%], and y=1.04x – 13.22 with R2 = 0.999. However, as shown in (d), Cartesian R2* appear to be no longer reliable beyond 200 mM of [Gd]. Note that the difference between (a) and (d) is the inclusion of R2* measurements from 225, 250, 275, and 300 mM of [Gd] to the linear regression analysis.

However, R2* measurements beyond 200 mM of [Gd] significantly affected the linear relationship for IDEAL-IQ (y = 5.85x + 127.61, R2 = 0.827) as opposed to cones (y = 10.87x – 166.96, R2 = 0.984) as shown in Figure 5 (d). The R2* measurements of IDEAL-IQ in these vials had larger SD than 3D multi-echo UTE cones (3 to 5 times) as listed follows. For the corresponding gadolinium vials (225, 250, 275, and 300 mM), mean values (and SD) of the ROI-based R2* measurements for 3D multi-echo UTE cones were 2301 (91.2), 2686 (95.2), 3055 (118.9), and 3399 (172.7) s−1. Mean values (and SD) of the ROI-based R2* measurements for IDEAL-IQ were 1565 (401.0), 1487 (481.0), 1029 (622.6), and 1805 (438.7) s−1. This is well appreciated in R2* maps associated with 225, 250, 275, and 300 mM of [Gd] in Figure 4, where noisier R2* is observed in IDEAL-IQ compared to 3D multi-echo UTE cones.

Reader Study – Image Quality Assessment

For Rater #1, the median scores of FB Cones, FB Cartesian, and BH Cartesian were 3 (IQR = 1), 2 (IQR = 2), and 4 (IQR = 1), respectively. For Rater #2, the median scores of FB Cones, FB Cartesian, and BH Cartesian were 4 (IQR = 1), 2 (IQR = 1), and 4 (IQR = 1), respectively. For each of the raters, p-value was <.001 from a one-sided Wilcoxon signed-rank test under the null hypothesis that the median score of FB Cones is less than or equal to the median score of FB Cartesian. P-value of <.001 was also calculated from a one-sided Wilcoxon signed-rank test performed under the null hypothesis that the median score of FB Cones is less than or equal to the median score of FB Cartesian and BH Cartesian combined. Confidence intervals for a population proportion of FB Cones to be preferred by each of the raters were calculated using a binomial exact calculation as shown in Figure 6. ICC was 0.788 with the 95% CI of [0.767, 0.808], which can be interpreted as good agreement (poor = <0.5; moderate = 0.5 to 0.75; good = 0.75 to 0.9; excellent > = 0.9). All the scores by the two raters are provided in Supporting Information Table S1.

Figure 6.

Figure 6.

Open in a new tab

Differences in scores with a graphical chart. The 95% CI that FB Cones (or simply Cones) will be preferred (difference of 1 or greater) by Reader #1 is 65-85% of the time. The 95% CI that Cones will be preferred by Reader #2 is 57-78% of the time. The 95% CI that BH/FB Cartesian (or simply Cartesian) will be preferred (difference of −1 or less) by Reader #1 is 5-20% of the time. The 95% CI that Cartesian will be preferred by Reader #2 is 2-14% of the time.

In Vivo R2* Comparison – Initial Dataset Screening

For quantitative comparison of ROI-based R2* between FB Cones and BH/FB Cartesian, 81 patient datasets were screened independently from the reader study for further analysis. 15 patient datasets out of 18 acquired with FB Cones and BH Cartesian were selected to proceed for further analysis. The excluded 3 patient datasets had 1 dataset of BH Cartesian with failed breath-holding and 2 datasets of FB Cones with bulk motion, such that ROI placement was not possible. Note that all 4 volunteer datasets were used. Similarly, 42 patient datasets out of 63 acquired with FB Cones and FB Cartesian were selected to proceed for further analysis. The excluded 21 patient datasets (33%, with 95% confidence interval of 23%-46%) had severe motion artifacts in the FB Cartesian R2* maps such that ROI placement was not possible. The artifacts in the FB Cartesian scans were likely caused by irregular breathing and bulk motion. Of note, the FB Cones R2* maps of these 21 patient datasets were not corrupted by severe motion artifacts so that ROI placement was possible.

In Vivo Study – Image Quality Scores and R2* measurements of Excluded/Included Datasets

The excluded datasets had a median score of 3 for FB Cones and 2 for BH/FB Cartesian combined. These scores were lower than the included datasets that had a median score of 4 for FB Cones and 3 for BH/FB Cartesian combined (4 for BH Cartesian and 3 for FB Cartesian). Among included datasets, 5 datasets had score 1 either for FB Cones or FB Cartesian: 1 dataset had FB Cartesian with score 1 from the both raters, 3 datasets had FB Cartesian with score 1 from one of the raters. Finally, 1 dataset had FB Cones with score 1 from one of the raters. A full list of image quality scores can be found in Supporting Information Table S1. The median R2* value of the excluded patients was 81 s−1 with the IQR of 12 s−1. R2* values of the excluded 3 BH Cartesian datasets were 24 s−1, 28 s−1, and 36 s−1. The median R2* values of the included patients who underwent FB Cones and FB Cartesian were 59 s−1 for FB Cones with the IQR of 125 s−1 and 57 s−1 for FB Cartesian with the IQR of 141 s−1. The median R2* values of the included patients who underwent FB Cones and BH Cartesian were 66 s−1 for FB Cones with the IQR of 210 s−1 and 79 s−1 for BH Cartesian with the IQR of 217 s−1.

In Vivo Study – Challenges in Cartesian CSE-MRI versus Addressed Challenges with FB Cones

Challenges in the conventional Cartesian CSE-MRI methods as exemplified in Figure 1 (a-c) were successfully addressed with FB Cones as shown in Figure 1 (d-f). Ghost artifacts along the phase encoding directions in Figure 1 (a, b) are no longer observable in Figure 1 (d, e). Image quality score of Figure 1 (a) was 2 by both Rater #1 and #2. Image quality scores of Figure 1 (b) was 2 by Rater #1 and 1 by Rater #2. Image quality score of Figure 1 (d) was 4 by both Rater #1 and #2. Image quality scores of Figure 1 (e) were 4 by Rater #1 and 5 by Rater #2. Compared to a large SD in R2* within the ROI shown in Figure 1 (c), a smaller SD in R2* was measured within the ROI shown in Figure 1 (f), suggesting better reliability in R2* measurements with FB Cones when the liver iron level is beyond mild overload, suggested by a FerriScan report (8.4 mg Fe/g dry tissue weight).

In Vivo Study – Representative R2* and ROI-based Quantitative Analysis

Although FB Cones and BH/FB Cartesian R2* had a similar visual appearance with median image quality scores of 3 to 4, some of the BH/FB Cartesian R2* maps had ghosting artifacts, as shown in Figure 7 (see also Supporting Information Videos S1 and S2) and Figure 8 (see also Supporting Information Videos S3 and S4). The top row of Figure 9 shows ROI-based Bland-Altman and linear correlation plots between FB Cones and BH Cartesian. The bias (mean difference, MD) was 2.23% and the SD was 6.59%. The 95% confidence interval (CI) of agreement limits (limits of agreement, LoA) was [−10.7%, 15.7%] across the range of measured R2* values, within which all the data points fell, suggesting a good agreement between FB Cones and BH Cartesian in ROI-based R2* measurements. A linear relationship was observed between BH Cartesian and FB Cones with y = 1.00x + 1.07 (R2 = 0.996). The bottom row of Figure 9 shows Bland-Altman and linear correlation plots between FB Cones and FB Cartesian. The bias was 0.21% and the SD was 7.02%. The LoA were [−13.5%, 14.0%], where all the measures fell, except for one data point with a relative difference of 20.5% between 53.5 s−1 of FB Cones R2* and 44.4 s−1 of FB Cartesian R2*, suggesting a good agreement between FB Cones and FB Cartesian in ROI-based R2* measurements. A linear relationship was also observed between FB Cartesian and FB Cones with y = 0.98x + 1.68 (R2 = 0.999).

Figure 7.

Figure 7.

Open in a new tab

Representative R2* of FB Cones and BH Cartesian in axial, coronal, and sagittal views with scan times (see also Supporting Information Videos S1 and S2). Two patients with suspected iron overload who underwent FB Cones and BH Cartesian are shown in (a) and (b), and (c) and (d). Image quality scores of (a), (b), (c), and (d) were (5 by Rater #1, 5 by Rater #2), (4, 4), (5, 5), and (5, 4), respectively. Although FB Cones and BH Cartesian had scores ≥ 4, mild ghosting artifacts are observable in the axial view of (b) and (d). Detailed imaging parameters are as follows: (a) no ACC, NEX=1, and res=2x2x3 mm3; (b) 4X ACC, NEX=0.75, and res=1.17x1.17x6 mm3; (c) no ACC, NEX=1, and rex=2x2x3 mm3; (d) 4X ACC, NEX=0.75, and res=1.56x1.56x6 mm3.

Figure 8.

Figure 8.

Open in a new tab

Representative R2* of FB Cones and FB Cartesian in axial, coronal, and sagittal views with scan times (see also Supporting Information Videos S3 and S4). Two patients with suspected iron overload who underwent FB Cones and FB Cartesian are shown in (a) and (b), and (c) and (d). Image quality scores of (a), (b), (c), and (d) were (5 by Rater #1, 5 by Rater #2), (4, 3), (4, 5), and (3, 4), respectively. As opposed to FB Cones, FB Cartesian exhibits ghosting artifacts in the axial view affecting R2* image quality as clearly seen in (b) and (d). Notice that a subtle heterogeneous iron distribution is observed in (a) with FB Cones as indicated by the red arrows. This heterogeneity was not faithfully captured in (b) with FB Cartesian because of ghosting artifacts. Detailed imaging parameters are as follows: (a) no ACC, NEX=1, and res=2.67x2.67x3 mm3; (b) no ACC, NEX=4, and res=1.88x1.88x6 mm3; (c) no ACC, NEX=1, and rex=2x2x3 mm3; (d) 2X ACC, NEX=4, and res=0.94x0.94x10 mm3.

Figure 9.

Figure 9.

Open in a new tab

In vivo ROI-based Bland-Altman and linear correlation plots. Top row shows comparisons of R2* [s−1] between FB Cones and BH Cartesian, where the MD is 2.23% and the SD is 6.59%, i.e., LoA = [−10.7%, 15.7%]. Bottom row shows comparisons of R2* [s−1] between FB Cones and FB Cartesian, where the MD is 0.21% and the SD is 7.02%, i.e., LoA = [−13.5%, 14.0%]. These ROI-based analyses suggest a good agreement between FB Cones and BH/FB Cartesian. The linear correlation plots show strong linear relationships between BH/FB Cartesian and FB Cones with R2 of 0.996 and 0.999, respectively.

DISCUSSION

We have developed a 3D multi-echo UTE cones MRI technique with golden-ratio re-ordering to enable motion-robust, fully ungated, and free-breathing R2* mapping of hepatic iron overload. The technique is based on CSE-MRI followed by a confounder-corrected R2* mapping method where the considered confounding factors are noise floor effects, B0 field inhomogeneity, and the presence of fat. The proposed technique has been demonstrated to show ability to capture rapid T2* decay via a phantom study and in vivo study on 3T clinical scanners. Supported by a reader study, the proposed FB Cones technique has superior motion robustness when compared to FB Cartesian strategies, with similar image quality to BH Cartesian imaging. ROI-based quantitative analyses performed on the visually eligible datasets showed an excellent agreement as well as a strong linear relationship between FB Cones and BH/FB Cartesian R2*.

With the growing use of non-Cartesian trajectories at 3T scanners, similar approaches have been recently proposed. A 2D multi-echo UTE imaging technique using center-out radial sampling studied by Krafft et al. 15 enabled free-breathing liver R2* quantification with 12 echoes and the acquisition time of < 2 min for a single 10 millimeter thick slice. This technique was evaluated by the same authors in sedated patients unable to perform BH maneuvers 16. For volumetric imaging, a center-out stack-of-stars radial technique with varying echo times 17 used a set of 7 single-echo images with 6 slices and 15 mm slice thickness, enabling free-breathing liver R2* quantification with the acquisition time of 35 sec. A 3D multi-echo stack-of-radial MRI technique that makes use of chemical shift encoding as well as the confounder-corrected R2* mapping18 was used for placenta R2* imaging with free breathing. This 3D multi-echo stack-of-radial MRI technique does not make use of center-out k-space sampling, such that the initial TE of 1.23 ms still remains as a challenge for R2* mapping of hepatic iron overload. As opposed to these methods using radial k-space trajectories, our 3D cones k-space trajectory inherently provides UTE imaging with higher sampling efficiency and greater motion robustness. Here, we have kept the cone readouts short to avoid off-resonance as a potential confounding factor for R2* quantification. We will explore re-designed long cone readouts in conjunction with off-resonance correction algorithms.

R2-based relaxometry can be also used for LIC estimation (FerriScan, Resonance Health Ltd). which has been only calibrated at 1.5T; this is a major drawback considering the growing popularity of clinical 3T scanners. There are a number of R2 spin echo sequences other than FerriScan with 5 to 20 min of imaging time 7; 19. Note that the subject shown in Figure 1 (c) and (f) was the patient who underwent both 3T and 1.5T to receive FerriScan (8.4 mg/g LIC). Since this patient is the only subject who received FerriScan and slice thickness of FB Cones for this subject was 6mm largely deviates from 3mm that we used for the 85 subjects we studied, this subject was excluded from our ROI analysis as robustness to imaging parameters has not been systematically investigated and is beyond the scope of this work.

Our phantom study assessed the capability of 3D multi-echo UTE cones MRI in capturing such a rapid T2* decay compared to IDEAL-IQ. The reported R2* measurements agreed with literature values 20, and the greater fidelity of multi-echo UTE cones in measuring high R2* likely resulted from a shorter initial TE. In our in vivo study, R2* maps captured heterogeneous iron distributions (Figure 8 and Supporting Information Video S3), potentially lost to liver biopsy or single-slice methods. Our in vivo reader study suggests that FB Cones R2* has better image quality than FB Cartesian, where ghost artifacts were still observed (Figure 8 and Supporting Information Videos S3 and S4), probably from irregular breathing. Although FB Cones R2* showed similar image quality to BH Cartesian (see also Supporting Information Table S1 for a full list of scores), BH Cartesian is still subject to motion artifacts (Figure 7 and Supporting Information Video S1). Yet, ROI-based analysis suggests a strong agreement between them, likely from averaging and highlighting the need for ROI-based analysis. An additional factor that may have caused visual differences between FB Cones and BH/FB Cartesian is the use of gradient nonlinearity distortion correction that was performed on the IDEAL-IQ sequence but not on FB Cones, but this barely affects R2* value. Lastly, the visual assessment performed via a reader study largely agreed with our data exclusions independently performed from the reader study as reported in the result. Agreement between the two raters was only fair, which should be further explored.

A recent study by Hernando et al. 21 showed that limits of agreement between two repeated IDEAL-IQ acquisitions at 3T were −16.6 to 15.5% across the range of R2* from 0 to 2000 s−1, considered to be highly repeatable in the context of hepatic iron assessment. The limits of agreement between FB Cones and BH/FB Cartesian were narrower than the repeatability study, suggesting a good agreement between these methods.

The noise performance of R2* mapping as a function of R2* value and TE combinations (#TE, initial TE, and echo spacing) has been extensively studied for Cartesian CSE-MRI 22; 23; 24. Our choice of TE combination for BH/FB Cartesian is similar to one of the investigated combinations in 22; 23; 24, where 6-8 echoes provided excellent SNR performance. Acquiring more echoes does not significantly improve R2* estimation but increases acquisition time. Our choice of TE combination for FB Cones is similar to BH/FB Cartesian in terms of #TEs and echo spacing; however due to the UTE component of FB Cones the TE range of FB Cones was shifted by the difference between the initial TEs of BH/FB Cartesian and FB Cones. Although a similar observation may be expected in regard of noise performance of R2* mapping with respect to TE combinations, investigating noise performance of our multi-echo UTE cones MRI would be an interesting future work.

Here, 46 patients underwent general anesthesia (GA) and the remaining 35 were awake. For Rater #1, the median score of GA patients was 3 with the IQR of 1 (average score 2.8, SD of 1.2). The median score of awake patients was also 3 with the IQR of 1 (average score 3.2, SD of 1.3). Similarly for Rater #2, the median scores of both GA and awake patients were 4 and the IQR was 1 (GA average 3.4 and SD 1.2, awake average 3.5 and SD 1.3). Further, out of 24 excluded patients 13 were awake and 11 were with GA. Therefore, we did not see any statistically significant differences between GA and non-GA patients in image quality score. A full list of GA/non-GA can be found in Supporting Information Table S1.

Limitations of this work are 1) a lack of comparison with a navigated or bellows-triggered IDEAL-IQ, 2) the large amount of excluded datasets (30%), 3) the phantom deviation from vivo signal (no fat, different relaxivities, etc), 4) moderate R2* range in patients, 5) blurring caused by bulk/respiratory motion, 6) different R2* mapping algorithms used in FB Cones and FB Cartesian, and 7) sensitivity to artifacts arising from MRI system imperfections such as eddy currents. Although navigated/bellows-triggered IDEAL-IQ has been proposed and used clinically, a comparative study was not carried out. At our institution, multi-NEX IDEAL-IQ (FB Cartesian) is clinically performed for liver R2* mapping when BH maneuver is not achievable. However, the proposed FB Cones has a clear echo time advantage; additionally, navigated IDEAL-IQ is still subject to some ghosting, which may affect R2* measurements.

Non-Cartesian trajectories have inherent advantages in terms of motion robustness, particularly reduced ghosting. Nevertheless, motion can still manifest as spatial blurring, affecting liver R2* measurements. Future work will focus on self-derived navigator signals to correct for respiratory motion25.

IDEAL-IQ product reconstruction was used for FB Cartesian since complex-valued source images were unavailable, perhaps causing the deviation in the Bland-Altman analysis (bottom row, Figure 9). IDEAL-IQ product reconstruction has limitations at high R2* 22, addressable partially by the complex nonlinear least-squares fitting used for FB Cones and BH Cartesian. However, within the R2* range in this study, no differences in R2* between these two reconstruction methods in BH Cartesian were observed. This may explain the MD of 2.23% between FB Cones and BH Cartesian (versus MD of 0.21% between FB Cones and FB Cartesian) where the same confounder-corrected R2* mapping algorithm was used for both can be addressed by motion-correction methods.

Eddy currents from time-varying gradients distort the nominal spatial encoding gradients and lead to trajectory errors that produce noticeable artifacts in non-Cartesian images. Thus, we approximated the eddy current effect in 3D cones as a constant time delay, empirically tuned to 3 us based on imaging a phantom, for all cone interleaves. However, this time delay should be re-calibrated for each scanner and is cone interleaf dependent 26. A fruitful direction for further work may be more comprehensive correction schemes based on impulse response characterization 27.

Future work will focus on multi-center validation and introducing both acceleration and self-navigation. Joint quantification of hepatic fat and iron 28 will also be explored. As shown in Table 1, the acquisition time of FB Cones was longer than FB Cartesian. Parallel imaging/compressed sensing (PICS) 29 can achieve two-fold acceleration. This will be extended to utilize sparsity along the echo dimension. Further, future work will leverage longer cone readouts for higher sampling efficiency enabling fully-sampled scans in significantly fewer TRs together with off-resonance correction 30. Lastly, a natural extension of the current work will be free-breathing liver QSM, studied preliminarily29; 31. The current acquisition and reconstruction pipeline generate high quality field maps which can be used for QSM.

CONCLUSION

3D multi-echo UTE cones MRI was developed and demonstrated to enable motion robust, fully ungated, and free-breathing R2* mapping of hepatic iron overload. For a visually eligible subset of the acquired data, FB Cones produced comparable R2* measurements and image quality to conventional BH Cartesian. Compared to FB Cartesian, FB Cones enabled R2* quantification in more patients and yields comparable R2* measurements with better image quality.

Supplementary Material

Video 1

Supporting Information Video S1. Axial slices of Figure 7 (a) and (b) are shown side-by-side. BH Cartesian is on the left and FB Cones is on the right. BH Cartesian had image quality score 4 from the both raters, and FB Cones had image quality score 5 from the both raters. These slices are played from the inferior to superior direction.

Download video file (11.3MB, mp4)
Video 2

Supporting Information Video S2. Axial slices of Figure 7 (c) and (d) are shown side-by-side. BH Cartesian is on the left and FB Cones is on the right. BH Cartesian had image quality scores 5 and 4 from each of the raters, and FB Cones had image quality score 5 from the both raters. These slices are played from the inferior to superior direction.

Download video file (6.4MB, mp4)
Video 3

Supporting Information Video S3. Axial slices of Figure 8 (a) and (b) are shown side-by-side. FB Cartesian is on the left and FB Cones is on the right. FB Cartesian had image quality scores 4 and 3 from each of the raters, and FB Cones had image quality score 5 from the both raters. These slices are played from the inferior to superior direction.

Download video file (7.6MB, mp4)
Video 4

Supporting Information Video S4. Axial slices of Figure 8 (c) and (d) are shown side-by-side. FB Cartesian is on the left and FB Cones is on the right. FB Cartesian had image quality scores 3 and 4 from each of the raters, and FB Cones had image quality scores 4 and 5 from each of the raters. These slices are played from the inferior to superior direction.

Download video file (10.6MB, mp4)
Table S1

Supporting Information Table S1. Image quality scores from two independent blinded raters (5-point scale: 1 = no liver can be evaluated; 2 = <1/3 of liver can be evaluated; 3 = 1/3 to 2/3 of liver can be evaluated; 4 = >2/3 of liver can be evaluated; 5 = all of liver can be evaluated).

Grant support:

NIH R01-DK117354, NIH R01-DK100651, NIH R01-EB026136, and GE Healthcare

REFERENCES

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Video 1

Supporting Information Video S1. Axial slices of Figure 7 (a) and (b) are shown side-by-side. BH Cartesian is on the left and FB Cones is on the right. BH Cartesian had image quality score 4 from the both raters, and FB Cones had image quality score 5 from the both raters. These slices are played from the inferior to superior direction.

Download video file (11.3MB, mp4)
Video 2

Supporting Information Video S2. Axial slices of Figure 7 (c) and (d) are shown side-by-side. BH Cartesian is on the left and FB Cones is on the right. BH Cartesian had image quality scores 5 and 4 from each of the raters, and FB Cones had image quality score 5 from the both raters. These slices are played from the inferior to superior direction.

Download video file (6.4MB, mp4)
Video 3

Supporting Information Video S3. Axial slices of Figure 8 (a) and (b) are shown side-by-side. FB Cartesian is on the left and FB Cones is on the right. FB Cartesian had image quality scores 4 and 3 from each of the raters, and FB Cones had image quality score 5 from the both raters. These slices are played from the inferior to superior direction.

Download video file (7.6MB, mp4)
Video 4

Supporting Information Video S4. Axial slices of Figure 8 (c) and (d) are shown side-by-side. FB Cartesian is on the left and FB Cones is on the right. FB Cartesian had image quality scores 3 and 4 from each of the raters, and FB Cones had image quality scores 4 and 5 from each of the raters. These slices are played from the inferior to superior direction.

Download video file (10.6MB, mp4)
Table S1

Supporting Information Table S1. Image quality scores from two independent blinded raters (5-point scale: 1 = no liver can be evaluated; 2 = <1/3 of liver can be evaluated; 3 = 1/3 to 2/3 of liver can be evaluated; 4 = >2/3 of liver can be evaluated; 5 = all of liver can be evaluated).

NIHMS1777036-supplement-Table_S1.pdf (54KB, pdf)

RESOURCES