Introduction

Nonalcoholic fatty liver disease (NAFLD) is the accumulation of hepatic fat in the absence of excessive alcohol intake when other liver disorders have been excluded. In the United States, 12.7 million children are obese [1] and 38% of obese children have NAFLD [2]. NAFLD can progress to nonalcoholic steatohepatitis, which can lead to cirrhosis, hepatocellular carcinoma and liver failure [3,4,5]. Liver biopsy is the gold standard for diagnosing and staging NAFLD [6, 7]. However, biopsies are invasive, limited by sampling bias, require anesthesia and can be associated with complications. Biopsies are also technically challenging [8, 9], particularly in obese children. Complicating matters, interobserver agreement among pathologists for NAFLD scoring can be variable [9,10,11]. For these reasons, there is a need for a noninvasive technique to accurately measure hepatic fat in children.

In adults, noninvasive magnetic resonance (MR) techniques have been developed to quantify hepatic fat. MR spectroscopy is regarded as a reference standard for noninvasive fat quantification [12, 13] and accurately quantifies hepatic steatosis [14,15,16]. However, MR spectroscopy only measures fat content in a single spatial location (i.e. voxel) and is limited by spatial sampling bias [12, 13]. Magnetic resonance imaging (MRI) is noninvasive and achieves spatially resolved hepatic fat quantification in 2-D slices or 3-D volumes by using chemical shift encoded (CSE) methods to calculate proton-density fat fraction (PDFF) [17,18,19]. To obtain accurate PDFF quantification, confounding factors such as complexities in the fat spectral model, T1 and T2* bias, gradient delay and eddy current errors, and noise bias must be considered [20]. CSE-MRI methods have mainly been investigated in adults with a few studies in children [19, 21,22,23]. In adults, CSE-MRI PDFF accurately detects and quantifies hepatic steatosis when compared to MR spectroscopy [24] and liver biopsy [25, 26].

Conventional MRI techniques, including CSE-MRI methods [24, 27,28,29], typically employ Cartesian trajectories for MR image acquisition [29,30,31]. Cartesian trajectories are limited by their sensitivity to respiratory motion-induced coherent aliasing artifacts, which degrade image quality and negatively impact PDFF quantification. In an attempt to compensate for this limitation, Cartesian MRI acquisitions of the liver are performed during a breath-hold. Because imaging parameters are selected to reduce scan times to fit within a breath-hold (typically 10 to 20 s), flexibility in acquiring larger spatial coverage, higher resolution, and/or greater signal-to-noise ratio (SNR) may be limited.

Additionally, a breath-hold is limited or not feasible in many patients (adults and children), specifically those with disabilities, neurological disorders, chronic illnesses, difficulties with communication or those unable to comply with operator instructions [29,30,31]. Even when children are able to suspend respiration, inconsistency and reduced breath-hold capacity compromise diagnostic information. While respiratory gating and self-navigation strategies for Cartesian MRI sequences can be used to mitigate motion artifacts, these strategies require longer and potentially variable scan times, and image quality may still suffer [29, 32, 33]. For these reasons, children may be sedated for abdominal MRI scans. However, sedation can have negative side effects and complications, particularly in children [31, 34]. As a result, there is a need for new free-breathing MRI techniques that overcome the challenges of traditional breath-hold MRI and avoid the need for sedation.

Non-Cartesian trajectories provide an alternative for MR imaging in children that eliminates the need for a breath-hold and sedation. Non-Cartesian 3-D stack-of-radial trajectories (also known as stack-of-stars) have dispersed motion artifacts in the radial encoding direction (in-plane) that do not obscure the anatomy, making radial trajectories more robust to breathing motion [29,30,31, 35,36,37]. As a result, radial MRI scans can be acquired during free-breathing, allowing for greater liver coverage, higher spatial resolution, and/or increased SNR. Recently, a new free-breathing multi-echo 3-D stack-of-radial MRI technique (free-breathing radial) with gradient error calibration and correction was developed for PDFF quantification in adults [36]. This technique acquires MRI data continuously during a free-breathing scan, reconstructs images and calculates PDFF maps. In adults, the hepatic PDFF measurements generated by free-breathing radial demonstrated a high degree of quantitative agreement to standard breath-hold MR spectroscopy and CSE-MRI techniques [36]. However, to date, there is no research on free-breathing hepatic fat quantification in children. Accordingly, this study’s purpose is to compare the image quality, accuracy and repeatability of this free-breathing radial technique [36] to conventional breath-hold MRI and breath-hold MR spectroscopy techniques for hepatic fat quantification in children.

Materials and methods

Pediatric study population

This study was approved by our local institutional review board. Parents/legal guardians provided informed consent and children provided assent. Healthy children and children with NAFLD, ages 7-17 years, were enrolled. Exclusion criteria included known liver disease (excluding NAFLD for NAFLD patients), infections, diseases or congenital anomalies known to affect the liver, contraindications to MRI and an inability to comply with breath-holding instructions.

Pediatric liver MR experiments

MRI experiments were performed on a 3-T scanner (MAGNETOM Skyra or Prisma; Siemens, Erlangen, Germany) using a spine array coil and body matrix array. Sequences acquired during the scan included a commercially available breath-holding 3-D Cartesian CSE-MRI sequence with controlled aliasing in parallel imaging results in higher acceleration (CAIPIRINHA) reconstruction (qDixon, the LiverLab package, software version VE11; Siemens, Erlangen, Germany), a commercially available breath-holding stimulated echo acquisition mode single-voxel MR spectroscopy (SVS) sequence (HISTO, the LiverLab package, software version VE11; Siemens, Erlangen, Germany), and a free-breathing radial sequence [28, 36, 39, 40]. Each sequence was repeated for each patient in random order within each scan session to evaluate repeatability. Except for the acceleration factor and the number of slices, the same imaging parameters were used for breath-holding Cartesian and free-breathing radial to enable a fair comparison between techniques (Table 1). The SVS voxel size was 25 mm × 25 mm × 25 mm with a total acquisition time of 15 s. SVS imaging parameters included echo times of 12 ms, 24 ms, 36 ms, 48 ms and 72 ms, a repetition time of 3,000 ms, a mixing time of 10 ms, a vector size of 1,024, and a bandwidth of 1,200 Hz/pixel. Additional MRI scans were repeated if motion artifacts resulted in nondiagnostic images. The breath-holding SVS region of interest (ROI) was positioned to avoid regions with image artifacts on breath-holding Cartesian, large blood vessels and bile ducts.

Table 1 Representative sequence parameters for in vivo pediatric liver experiments
Full size table

MR image reconstruction and PDFF calculation

Breath-holding Cartesian acquisitions were reconstructed and PDFF maps were determined by a prototype scanner software (Work-In-Progress package 963, software version VE11; Siemens, Erlangen, Germany) [28]. The breath-holding Cartesian PDFF maps were calculated using a 7-peak fat model and single effective R2* per voxel; breath-holding SVS PDFF was calculated by the scanner software using T2 correction [28, 40, 41]. Free-breathing radial data sets (fully sampled based on Nyquist criteria) were reconstructed, without discarding data or employing parallel imaging reconstruction techniques. Free-breathing radial PDFF maps were calculated offline in MATLAB R2016a (MathWorks, Natwick, MA, USA) using the same signal model used for breath-holding Cartesian PDFF maps, complex fitting with a graph cut algorithm, and magnitude discrimination [36, 41,42,43,44,45].

Image analysis

All reconstructed images and PDFF maps were converted to DICOM (Digital Imaging and Communications in Medicine) for analysis in OsiriX software version 6.0 (Pixmeo Sarl, Bernex, Switzerland). Image quality was evaluated by an experienced pediatric radiologist (S.G., >10 years of experience) masked to the trajectory (Cartesian or radial) and patient, by scoring the first echo time images (TE=1.23 ms) and PDFF maps in axial and coronal orientations. Images were scored on a scale of 1-4 for motion artifacts, other imaging artifacts and image blurring. Motion artifacts included coherent aliasing artifacts on breath-holding Cartesian images and radial streaking on free-breathing radial images. Other artifacts included CAIPIRINHA reconstruction artifacts on breath-holding Cartesian images and any additional non-motion-related artifacts on breath-holding Cartesian and free-breathing radial images. Motion blurring referred to the blurring of blood vessels, abdominal wall or liver dome on breath-holding Cartesian and free-breathing radial images. The criteria for a score of 1 were nondiagnostic images with significant artifacts that confound PDFF (i.e. bad image quality); 2 indicated diagnostic images with artifacts that confound PDFF, 3 indicated diagnostic images with artifacts that did not confound PDFF and 4 indicated no artifacts (i.e. good image quality). The percentage of images that fell into each score category was determined. To characterize liver coverage, the number of slices that contained the liver using the breath-holding Cartesian scan (NslC) and the free-breathing radial scan (NslR) was counted. Free-breathing radial scans were always prescribed to cover the entire liver. The liver slice coverage (Lsl) was then calculated as the percentage of liver slices covered using breath-holding Cartesian relative to free-breathing radial (i.e. Lsl=NslC/NslR×100%) and reported as mean±standard deviation (SD). To evaluate quantitative hepatic PDFF accuracy, one ROI was drawn on each of the breath-holding Cartesian and free-breathing radial PDFF maps in an anatomical location corresponding to the breath-holding SVS ROI that was placed during the MR scan, and PDFF was recorded as mean±SD.

Statistical analysis

Age was described for healthy subjects and patients with NAFLD and the median values and interquartile range (IQR) were compared using a nonparametric two group Mann-Whitney U test due to the limited sample size [46, 47]. Paired differences in image quality scores between breath-holding Cartesian (i.e. conventional reference) and free-breathing radial scans, were assessed using McNemar-Bowker tests for dependent categorical data to compare the distribution of scores by cohort (healthy cohort vs. NAFLD cohort) and artifact category (motion artifacts, other artifacts and motion blurring) [48].

Simple linear regression analysis was performed to compare techniques by determining the equation for the linear relationship, Pearson’s correlation coefficient (r) and Lin’s concordance correlation coefficient (ρc) [49, 50]. Pearson’s r was calculated to evaluate the strength of the linear associations between the two techniques, whereas Lin’s ρc was calculated to estimate the degree of quantitative agreement (i.e. to estimate whether free-breathing radial or breath-holding techniques produce the exact same value for the range of hepatic PDFF). For example, when two techniques produce the exact same hepatic PDFF, ρc is 1. Lin’s ρc was reported in this study because a high Pearson’s r does not necessarily imply identical measurements between free-breathing radial and breath-holding techniques. The equation for the concordance coefficient comparing free-breathing radial to breath-holding techniques is shown in the Appendix. Bland-Altman analysis was performed to assess differences between two techniques by plotting the PDFF difference between the two techniques against the mean PDFF between the two techniques [51]. Specifically, the Bland-Altman analysis measures the mean difference (MD) or bias between two methods and the 95% limits of agreement (LoA), which were reported as mean difference (MD)±1.96 times the SD (i.e. ±LoA=MD±1.96×SD).

To assess repeatability for each technique, the within-subject mean difference (MDwithin), within-subject SD (SDwithin) and within-subject coefficient of repeatability (CRwithin) between repeated scans were reported [52, 53]. The CRwithin measures the variability for repeated measurements of the same technique (i.e. CRwithin=1.96\( \times \sqrt{2}\times \)SDwithin), therefore a smaller CRwithin demonstrates smaller differences between repeated measurements [53]. Statistical analyses were performed in STATA software version 12.0 (Statacorp, College Station, TX) and MATLAB. A P-value <0.05 was considered significant.

Results

Pediatric study population

Ten healthy subjects and ten patients with NAFLD were enrolled. Of these, all ten healthy subjects (6 male, median age [±IQR]: 10.9 [±3.3] years) and 9 NAFLD patients (7 male, age: 15.2 [±3.2] years) completed the study. The median age between the two groups was significantly different (P<0.05). One patient with NAFLD did not complete the study because the MRI field of view was not adequate to cover his/her body habitus. All of the patients with NAFLD were confirmed by breath-holding SVS or breath-holding Cartesian to have hepatic PDFF >5.6%, consistent with a diagnosis of NAFLD [14, 54].

Image quality

If the breath-holding Cartesian images showed significant aliasing artifacts due to motion or if the breath-holding liver position changed such that the images from the breath-holding Cartesian scan did not adequately cover the liver volume, breath-holding Cartesian scan(s) were repeated. For some subjects, motion-induced aliasing or other artifacts were still observed on breath-holding Cartesian images even when repeated. The number of repeated breath-holding Cartesian scans ranged from 0 to 2 for all subjects. Specifically, the mean number of repeated breath-holding Cartesian scans performed was 0.40±0.70 in healthy subjects and 0.89±0.93 in patients with NAFLD. In contrast, the free-breathing radial scan was only repeated once for one healthy subject.

The free-breathing radial technique demonstrated robustness to motion with good image quality and quantitative PDFF maps. Representative images of breath-holding Cartesian and free-breathing radial acquisitions with motion artifact image quality scores ranging from 1 to 4 are shown in Fig. 1. Figure 2 shows an example of breath-holding Cartesian and free-breathing radial images and PDFF maps from a representative subject where major and minor motion occurred during a breath-holding Cartesian scan.

Fig. 1
figure 1

Representative images in axial and coronal reformat orientations at an echo time of 1.23 ms with motion artifact scores of 1-4 for (a) breath-holding Cartesian and 2-4 for (b) free-breathing radial acquisitions. If the breath-holding Cartesian acquisition exhibited severe motion artifacts leading to nondiagnostic images (score of 1), additional scans were repeated during the scan session. The free-breathing radial acquisition did not have a representative image with an image quality score of 1. See Materials and methods for the description of the image quality scores

Full size image
Fig. 2
figure 2

Images and proton-density fat fraction (PDFF) maps for the free-breathing radial and breath-holding Cartesian scans for a representative NAFLD patient (15-year-old boy) in axial orientation. Due to severe motion artifacts (nondiagnostic image quality) in the breath-holding Cartesian scan (motion artifact score=1), it was repeated. However, mild motion artifacts (motion artifact score=3) were still present in the repeated scan. Arrows indicate artifacts that affect hepatic fat quantification. The free-breathing radial scan did not exhibit coherent motion artifacts in the liver (motion artifact score=3). The scan time for each technique is reported as minutes:seconds

Full size image

Axial, reformatted coronal and reformatted sagittal PDFF maps from free-breathing radial and breath-holding Cartesian techniques for a representative healthy subject and patient with NAFLD are shown in Figs. 3 and 4, respectively. The PDFF maps from free-breathing radial scans and breath-holding Cartesian scans are similar, but have slight differences in liver position due to the breath-holding. Additionally, in some NAFLD patients hepatic PDFF heterogeneity was observed using both breath-holding Cartesian and free-breathing radial techniques (Fig. 5).

Fig. 3
figure 3

Proton-density fat fraction (PDFF) maps for breath-holding Cartesian and free-breathing radial scans for a representative healthy subject (9-year-old boy) in axial, coronal and sagittal orientations. The liver slice coverage was 100% using the breath-holding Cartesian technique. Representative regions of interest (boxes) and mean PDFF values are shown in the axial orientation. Free-breathing radial and breath-holding Cartesian scans have slight differences in liver position due to breath-holding. The scan time for each technique is reported as minutes:seconds

Full size image
Fig. 4
figure 4

Proton-density fat fraction (PDFF) maps for breath-holding Cartesian and free-breathing radial scans for a representative nonalcoholic fatty liver disease (NAFLD) patient (14-year-old girl) in axial, coronal and sagittal orientations. The liver slice coverage was 68% using the breath-holding Cartesian technique. Representative regions of interest (boxes) and mean PDFF values are shown in the axial orientation. Free-breathing radial and breath-holding Cartesian scans have slight differences in liver position due to breath-holding. The scan time for each technique is reported as minutes:seconds

Full size image
Fig. 5
figure 5

Proton-density fat fraction (PDFF) maps for breath-holding Cartesian and free-breathing radial scans for a representative nonalcoholic fatty liver disease (NAFLD) patient (10-year-old boy) with hepatic PDFF heterogeneity in axial, coronal and sagittal orientations. The liver slice coverage was 97% using the breath-holding Cartesian technique. Representative regions of interest (circles) and mean PDFF values are shown in three liver segments. Free-breathing radial and breath-holding Cartesian scans have slight differences in liver position due to breath-holding. The scan time for each technique is reported as minutes:seconds

Full size image

In healthy subjects, the breath-holding Cartesian scan achieved full liver slice coverage with the resolution used in this study. However, in most patients with NAFLD, full liver slice coverage was not achieved using the breath-holding Cartesian technique. In patients with NAFLD, the liver slice coverage was 100%±0% for free-breathing radial scans while it was 74%±17% for breath-holding Cartesian scans. To maintain similar resolution among all subjects, many NAFLD patients required a larger matrix size than healthy subjects due to a larger field of view. Furthermore, some patients with NAFLD had limited breath-holding ability and the volumetric coverage needed to be reduced further to decrease the breath-holding time. Table 2 shows that for patients with NAFLD, free-breathing radial scans had a significantly higher proportion of good motion artifact and other artifact image quality scores compared to breath-holding Cartesian scans (P<0.05). In both patients with NAFLD and healthy subjects, the proportion of image blurring scores was not significantly different between free-breathing radial scans and breath-holding Cartesian scans.

Table 2 Radiologist image quality scores for the breath-holding (BH) Cartesian and free-breathing (FB) radial first echo time images and proton density fat fraction (PDFF) maps for motion artifacts, other artifacts, and image blurring for the healthy pediatric subjects and nonalcoholic fatty liver disease (NAFLD) pediatric patients. Images were scored on a scale of 1–4 where a score of 1 indicates bad image quality and 4 indicates good image quality
Full size table

Hepatic PDFF quantification accuracy

To avoid PDFF errors due to coherent Cartesian aliasing artifacts, all ROIs measured using the breath-holding SVS, free-breathing radial images, and breath-holding Cartesian images were positioned away from Cartesian MRI artifacts for quantitative analysis. The results showed a linear relationship between breath-holding Cartesian PDFF, breath-holding SVS PDFF and the proposed free-breathing radial PDFF. All comparisons had a significant r and ρc >0.99 (P<0.001) and MD <0.7%. The linear correlation and Bland-Altman plots are shown in Fig. 6. The comparison between the proposed free-breathing radial PDFF versus breath-holding Cartesian PDFF showed a slope of 1.03 for the linear regression and correlation coefficients r and ρc were 0.996 and 0.994, respectively. For the proposed free-breathing radial PDFF versus breath-holding SVS PDFF, the slope for the linear regression was 0.97 and the correlation coefficients r and ρc were 0.997 and 0.995, respectively. The conventional breath-holding Cartesian PDFF versus the reference breath-holding SVS PDFF comparison showed a slope for the linear regression of 0.94 and the correlation coefficients r and ρc were 0.997 and 0.995, respectively. The mean difference between free-breathing radial PDFF and conventional breath-holding Cartesian PDFF was 0.65%±2.56%, the mean difference between free-breathing radial PDFF and reference breath-holding SVS PDFF was 0.64%±2.31%, and the mean difference between conventional breath-holding Cartesian PDFF and reference breath-holding SVS PDFF was 0.23%±2.56%. Free-breathing radial PDFF, breath-holding Cartesian PDFF and breath-holding SVS PDFF demonstrated repeatability with MDwithin=0.25%, -0.09% and -0.30%, respectively; SDwithin=0.55%, 0.32% and 1.35%, respectively; CRwithin=1.53%, 0.89% and 3.74%, respectively. A summary of the repeatability analysis is shown in Table 3.

Fig. 6
figure 6

Liver study (a, c, e) linear correlation plots and (b, d, f) Bland-Altman plots for proton-density fat fraction (PDFF) of each region of interest (ROI) in the liver corresponding to the single-voxel magnetic resonance spectroscopy (SVS) ROI. The comparison of (a-b) free-breathing radial versus breath-holding Cartesian had mean difference =0.65%±2.56% (c-d) free-breathing radial versus breath-holding SVS had mean difference =0.64%±2.31%, and (e-f) breath-holding Cartesian versus breath-holding SVS had mean difference=0.23%±2.56%. The correlation coefficients r and ρc were significant in all cases with P<0.001. The dashed lines represent y=x in the linear correlation plots and y=0 in the Bland-Altman plots. BH breath-holding, FB free-breathing

Full size image
Table 3 The analysis of repeatability results. The within-subject mean difference (MDwithin), within-subject standard deviation (SDwithin), and within-subject coefficient of repeatability (CRwithin) are reported between repeated scans for the breath-holding (BH) single-voxel magnetic resonance spectroscopy (SVS), BH Cartesian and free-breathing (FB) radial techniques
Full size table

Discussion

In this study, we evaluated the accuracy and repeatability of 3-D hepatic PDFF quantification using a free-breathing radial MRI technique with respect to established breath-holding techniques in a pediatric population. This study provides evidence that accurate and repeatable free-breathing hepatic fat quantification can be performed in children for the entire 3-D liver volume using a multi-echo 3-D stack-of-radial technique. Hepatic PDFF calculated from free-breathing radial scans had significant linear correlation, significant concordance and low mean differences <0.7% compared to the conventional breath-holding Cartesian and the reference standard breath-holding SVS techniques. Free-breathing radial, breath-holding Cartesian and breath-holding SVS techniques demonstrated repeatability with MDwithin<0.3%. Free-breathing radial scans and breath-holding Cartesian scans demonstrated repeatability with CRwithin<2%. Breath-holding SVS scans had a larger CRwithin of 3.74% likely due to variation in the breath-holding position and actual voxel position between breath-holding SVS scans. While the mean PDFF differences between free-breathing radial and breath-holding techniques were very small, they may reflect differences due to the breath-holding position, SVS partial volume effects and the effect of motion on the images. Importantly, in patients with NAFLD, free-breathing radial scans demonstrated a significantly higher proportion of good image quality scores compared with breath-holding Cartesian scans. We speculate that overweight patients with NAFLD may have had more difficulty performing a breath-hold than their healthy peers. The median (±IQR) body mass index in the NAFLD cohort was 33.2 (±7.1) kg/m2. In contrast, the body mass index in the healthy cohort was normal at 17.4 (±2.0) kg/m2. Therefore, compared to the conventional breath-holding Cartesian technique, free-breathing radial scans may provide improved image quality and hepatic PDFF quantification for evaluating steatosis in this target population.

There have been a few studies evaluating breath-holding Cartesian CSE-MRI techniques for hepatic PDFF quantification in children [19, 21,22,23]. Two studies used dual-echo and triple-echo methods for PDFF quantification [21, 23]. However, dual-echo and triple-echo approaches do not correct for all confounding factors (such as T2* bias and the multi-peak spectrum of fat) and, as a result, hepatic PDFF quantification may be inaccurate [20]. Since the signal model includes many parameters, at least four echoes are recommended to account for confounding factors such as T2* bias and the multi-peak spectrum of fat (e.g., not correcting for T2* can cause an underestimated hepatic fat fraction) [20, 55]. Other studies utilized a six-echo technique with magnitude-based fitting for PDFF estimation in the liver and found agreement between breath-holding Cartesian PDFF and breath-holding SVS PDFF, correlation between breath-holding Cartesian PDFF and histopathology, and associations between breath-holding Cartesian PDFF and clinical characteristics [19, 22, 23]. These studies employed a 2-D breath-holding Cartesian MRI technique, which may have limited spatial coverage or resolution and motion artifacts [19, 21,22,23].

In this study, variation in the distribution of image quality scores between breath-holding Cartesian and free-breathing radial was dependent upon the artifact category and subject population. Blurring was observed in breath-holding Cartesian and free-breathing radial images with no statistically significant differences in the distribution of scores for both healthy subjects and patients with NAFLD. This did not affect PDFF quantification. In the motion artifact category, breath-holding Cartesian images had coherent aliasing artifacts that may confound hepatic PDFF while free-breathing radial images had radial streaking artifacts that interfere less with hepatic PDFF. In patients with NAFLD, the percentage of bad image quality scores for motion artifacts was higher for the breath-holding Cartesian technique than for the free-breathing radial technique. We believe this is because children with NAFLD may have difficulty performing a breath-hold due to impaired lung function secondary to their overweight/obese status. Additionally, in patients with NAFLD, breath-holding Cartesian scans showed a higher percentage of bad image quality scores for other artifacts. One of the main sources of these artifacts is related to undersampling and CAIPIRINHA reconstruction errors. This is most likely due to the fact that the center of the field of view’s coil sensitivity is reduced by a large body size. In order to improve breath-holding Cartesian image quality for patients with NAFLD, less undersampling may be required, which would increase scan time, reduce volumetric coverage and/or reduce spatial resolution. On the other hand, the free-breathing radial technique does not necessarily require parallel imaging or have breath-holding scan time limitations. Therefore, image quality can be improved by allowing for longer free-breathing acquisition times for patients with NAFLD.

In this study, the breath-holding Cartesian technique required four-fold undersampling to reduce the scan time to within a breath-hold. For this reason, volumetric coverage was limited. As a result, for children with hepatomegaly, full liver slice coverage with the desired spatial resolution may not be achievable within a breath-hold. Free-breathing radial scans can achieve greater volumetric coverage and/or spatial resolution compared to breath-holding Cartesian scans, thereby improving 3-D spatial characterization of hepatic PDFF and liver disease. This may be important for cases where hepatic PDFF is heterogeneous (Fig. 5). Other studies have reported heterogeneous hepatic PDFF in adults and children, and recent evidence suggests heterogeneity in liver fat reduction following bariatric surgery [56,58,59,60,60]. However, additional studies are required for a full understanding of the spatial distribution of PDFF in patients with NAFLD. In addition, full liver coverage achieved by free-breathing radial MRI may have additional clinical applications including the evaluation of hepatic masses or other diseases. Traditionally, complex hepatic diseases require sedation or general anesthesia for MRI in order ensure full liver coverage and adequate resolution. Free-breathing radial MRI technology eliminates this requirement, and at the same time can yield diagnostically useful information. Each free-breathing radial acquisition in this study was performed with the same image resolution for children with either a small or large body habitus, and the number of slices was chosen to cover the entire liver volume. For these reasons, fully sampled free-breathing radial acquisitions were approximately 2-5 min depending upon the subject’s size (Table 1). The free-breathing radial scan time could be shortened if fewer slices are acquired or a lower resolution is used. On the other hand, higher resolutions may improve diagnostic information when performing MRI scans, but this may require longer scan times or reduced liver slice coverage. While the radial trajectory was fully sampled in this study, radial undersampling can be performed in combination with non-Cartesian parallel imaging reconstruction [61]. Previous work in adults showed similar PDFF quantification results using an undersampled free-breathing radial technique, potentially reducing the scan time to less than 1.5 min [36].

In addition to the constraints of the breath-holding Cartesian technique with respect to liver coverage and spatial resolution, patient motion leads to artifacts and reduces image quality. Previous studies have developed respiratory gating and self-navigation strategies for hepatic fat quantification using Cartesian MRI sequences to reduce motion artifacts in adults [32, 33]. To our knowledge, these strategies have not been evaluated for hepatic fat quantification in children. In adults, respiratory gating and self-navigation have shown similar image quality and fat quantification accuracy compared to breath-holding techniques [32, 33]. However, these strategies require longer scan times for image acquisition and are well-suited for patients with regular respiratory patterns, but may not be appropriate for children who require shorter scan times and have heavy or irregular breathing [29, 33]. In fact, respiratory gating and self-navigation strategies may benefit from the incorporation of non-Cartesian trajectories, such as 3-D stack-of-radial, to accommodate patients with heavy or irregular breathing. Although self-navigation using the free-breathing radial technique was not evaluated in this current study, the 3-D stack-of-radial trajectory allows for self-navigation and improved image quality in the presence of motion [38, 62].

The signal model used in this study includes fitting for R2*, which can be used to assess hepatic iron content [63,66,67,66]. Severe hepatic iron overload may hinder PDFF quantification due to rapid signal decay prior to acquiring the first echo time [63]. To address this, recent work includes ultra-short echo time sequences for R2* mapping in patients with severe iron overload [67,68,69]. In this study, there were no subjects with severe iron overload. Therefore, hepatic PDFF quantification was not confounded by high R2* values.

Our study has limitations. First, partial-volume effects and motion can confound breath-holding SVS scans, which were not corrected. Second, inter-sequence motion and non-rigid changes in liver tissue during breath-holding and free-breathing may hinder PDFF comparisons between techniques. To address this limitation, ROIs were placed in corresponding anatomical locations on breath-holding Cartesian and free-breathing radial PDFF maps. Third, the scan time for the radial acquisition is increased by 31-45 s due to the addition of a calibration scan for radial MRI, and the number of radial readouts needed for calibration was not optimized to reduce scan time. Finally, the sample size is small and this study was performed at a single site. However, the results of this study indicate that free-breathing radial is not only comparable to conventional breath-holding MR techniques, but also has some additional advantages for hepatic fat quantification in children.

Conclusion

In this study of healthy children and children with NAFLD, free-breathing hepatic fat quantification using a new 3-D stack-of-radial MRI technique was accurate and repeatable. The free-breathing radial technique demonstrated significantly less image artifacts than breath-holding Cartesian scans in patients with NAFLD. These results show that the free-breathing radial technique may potentially improve pediatric patient compliance and 3-D spatially resolved characterization of hepatic steatosis. Moreover, in this study, free-breathing radial eliminated the need for breath-holding and did not require sedation for fat quantification. The free-breathing radial technique represents a promising tool that could help clinicians diagnose and manage children with NAFLD.