Comparison of Volumetric Measurement Method With Region of Interest Drawing Method for Liver Fat Quantification

Nonalcoholic fatty liver disease (NAFLD) occurs when fat accumulates in the liver without a specific cause. The disease presents with varying degrees of severity and includes simple steatosis, nonalcoholic steatohepatitis (NASH), hepatic fibrosis, liver cirrhosis, and hepatocellular carcinoma (HCC) [1]. NAFLD is the most prevalent chronic liver disease in Western countries, affecting 30% of the general population [2]. In addition to morbidity and mortality, NAFLD is closely associated with the development of metabolic syndromes, including type 2 diabetes, obesity, dyslipidemia, and atherosclerotic cardiovascular disease [3, 4, 5]. For this reason, significant effort has been devoted to understanding the pathogenesis of NAFLD, and novel therapies based on these new findings are being tested in clinical trials [6].

Assessment of hepatic fat deposition is required for both the development of new therapies and follow-up of patients with NAFLD. Traditionally, a liver biopsy has been considered the reference standard for diagnosing liver fat deposition. However, performing a liver biopsy for fat analysis has limitations such as difficulty in repetition, sampling errors, and complications including bleeding [7, 8, 9]. As a result, there is an unmet need for non-invasive methods to evaluate hepatic steatosis, and imaging modalities have been implemented for this purpose. Ultrasound (US) and computed tomography (CT) have been used to evaluate hepatic steatosis. However, US is highly operator dependent, and the clinical implementation of US methods for hepatic steatosis assessment may be limited by a variety of vendor-specific techniques and cut-off values [10, 11, 12]. CT exposes patients to radiation and has limited sensitivity and specificity in diagnosing mild fatty liver disease [13, 14, 15]. In addition, various conditions such as superimposed iron, amiodarone, iodine contrast, glycogen overload, and hepatitis can act as confounding factors that increase liver attenuation [15]. Recently, the controlled attenuation parameter (CAP) has been used for the overall assessment of hepatic steatosis owing to its noninvasiveness and good diagnostic performance. Furthermore, with the development of the XL probe, hepatic steatosis can be accurately diagnosed even in obese patients. However, its use is still limited in patients with bowel interposition or masses in the right hepatic lobe. CAP values may also be affected by body mass index, the presence of diabetes mellitus or NAFLD/NASH, and even the probe used (M or XL) [16].

Hepatic steatosis can be accurately evaluated by magnetic resonance imaging (MRI). MRI-based fat quantification methods are robust and reliable because they are not affected by scanner characteristics, acquisition parameters, or field strength [17, 18]. Furthermore, the method presents no concerns regarding radiation exposure and patient conditions such as ascites or obesity. The proton density fat fraction (PDFF) is defined as the ratio of unconfounded signals from the mobile protons of fat to unconfounded signals from the mobile protons of water [14, 17]. The PDFF is optimal for screening, diagnosing, and monitoring hepatic steatosis and is already widely accepted as a quantitative biomarker of hepatic steatosis [18, 19]. Recently, owing to the complexity of both the acquisition and processing processes as well as limited spatial coverage, chemical shift–encoded MRI has been more widely performed than magnetic resonance spectroscopy [14]. However, there is currently no consensus regarding the best approach for deriving summary quantitative PDFF estimates from PDFF maps acquired across the whole liver [14]. Considering the heterogeneous nature of hepatic fat deposition, a method for measuring liver fat by drawing multiple large regions of interest (ROIs) was introduced [20, 21, 22]. Measurement using the 4-ROI (medial, lateral, anterior, and posterior) or 9-ROI paradigm (one in each of the Couinaud segments, including 4a and 4b) is generally recommended, but known limitations include the long time required for measurement, sensitivity to imaging artifacts, and inaccurate representation of PDFF if there is focal or regional fat deposition/sparing [23, 24, 25]. Recently, a commercially available program (Philips IntelliSpace Portal Client, Philips Healthcare, Best, The Netherlands) has provided volumetric measurement of PDFF through automated whole-liver segmentation. Although attempts have been made to take volumetric measurements of hepatic steatosis using in-house software or commercial software, previous studies have focused more on the accuracy of volumetric segmentation of the liver than on comparing the different methods available to measure fat by drawing ROIs, or did not demonstrate how much time could be reduced by using volumetric fat measurement compared to conventional methods [26, 27, 28]. In this study, in addition to comparing the liver fat measurement results, we measured the time required to perform fat quantification to confirm the efficacy of automated whole-liver segmentation. Hence, the purpose of this study was to investigate the feasibility of a volumetric liver fat measurement method (Fat_vol) compared with the conventional 27-regions of interest drawing method (Fat_roi).

This retrospective study was approved by the Institutional Review Board of Severance Hospital (No.4-2021-0290), which waived the requirement for informed consent.

MRI Post-Processing and Analysis

PDFF was measured independently by two radiologists (reviewer A, a senior radiological resident, and reviewer B, a radiologist with 4 years of experience in abdominal radiology) using two methods: Fat_roi and Fat_vol. For the Fat_roi method, the reviewers drew ROIs as large as possible in each of the Couinaud segments (including separate measurements for 4a and 4b) while avoiding major blood vessels, bile ducts, and artifacts on the fat map. Three ROIs were drawn per segment in different slices and a total of 27 values were acquired (Fig. 1). The measured values were recorded using a spreadsheet program, and the average PDFF was automatically calculated according to a pre-entered formula. Additionally, the total time from the start of drawing the first ROI to recording the final PDFF value was measured. For the Fat_vol method, PDFF was measured using commercially available software (Liver Health, IntelliSpace Portal Client, Philips Healthcare, Best, The Netherlands). First, the liver was automatically segmented, and the program presented axial, coronal, and sagittal views of the segmented area. Subsequently, the reviewers corrected the liver margin if unnecessary parts such as the gallbladder, major blood vessels, or other adjacent organs were included or if the liver parenchyma was not included. However, as exact segmentation along the liver boundary is time-consuming, the reviewers were asked to exclude structures other than the liver parenchyma (such as the gallbladder or heart) that have similar signal intensity to the liver, but to leave the boundary line drawn slightly inside the liver margin, as initially presented. This choice was made because the purpose of this study was not to accurately segment the liver but to evaluate the clinical feasibility of Fat_vol. Excluding a minimal volume of the liver parenchyma along the margin was not expected to significantly affect the final results (i.e., the volumetric PDFF value), whereas the inclusion of other structures was expected to have a large impact because fat is barely deposited in other organs. After confirming correct segmentation of the liver, the volumetric PDFF was calculated using a histogram of the calculated values. To remove noise (or error) and structures other than the liver within the volumetric ROI, the lower fat fraction margin was set to 0%, and the upper fat fraction margin was set based on the graph showing a normal distribution. After displaying the data in a histogram to obtain an appropriate value for liver fat quantification, the average value and standard deviation of liver fat were recorded (Fig. 2). The time from the start of the volumetric measurement to the recording of the PDFF value was also measured. Both reviewers were asked to measure the PDFF twice using each method to evaluate intra-observer variability. To prevent recall bias, measurements using the same method were performed with a time interval of 4–8 weeks.

Fig. 1
Multiple Fat_roi. Three ROIs are drawn in each of nine segments of the liver for fat quantification. Fat_roi, 27-regions of interest drawing method; ROI, regions of interest.

• Click for larger image
• Download as PowerPoint slide

Fig. 2
Volumetric liver fat measurement method (Fat_vol). Automatic liver segmentation is conducted on axial (A), coronal (B) and saggital (C) images, excluding organs such as the gallbladder, bowels, and stomach, for fat quantification. The result is presented as a 3D image (D).

• Click for larger image
• Download as PowerPoint slide

Statistical Analysis

The intra-class correlation coefficient (ICC) was obtained for intra-exam repeatability, inter-reviewer reproducibility, and inter-exam agreement (Fig. 3). The one-way random method was used for intra-exam repeatability. The two-way random method was used to assess inter-reviewer reproducibility. The two-way mixed method was used to determine inter-exam agreement. ICC values greater than 0.9 were considered to indicate excellent agreement. For ICC values between 0.75 and 0.90, the agreement was considered good, those between 0.50 and 0.75 were considered moderate, and those below 0.50, poor. The mean differences in measurement time within the reviewers and PDFFs were compared using a paired t-test. Bland–Altman analysis was used to assess intra-exam repeatability, inter-reviewer reproducibility, and inter-exam agreement. In addition, the correlation between Fat_roi and Fat_vol measured by both reviewers was evaluated using Pearson’s correlation coefficient. The interpretation of the correlation coefficient was as follows: weak, 0–0.4; moderate, 0.4–0.7; and strong, 0.7–1. Statistical significance was set at p < 0.05. All statistical analyses were performed using SPSS version 25.0 (IBM Corp., Armonk, NY, USA).

Fig. 3
Intra-exam repeatability (red arrows), inter-reviewer reproducibility (yellow arrows), and inter-exam agreement (purple arrows) analyses for conventional 27-ROI method (Fat_roi) and volumetric method (Fat_vol). ROI, regions of interest; Fat_roi, 27-regions of interest drawing method; Fat_vol, volumetric liver fat measurement method.

• Click for larger image
• Download as PowerPoint slide

The results of this study demonstrate that use of the Fat_vol method could reduce liver fat measurement times significantly compared to the conventional Fat_roi method while preserving inter- and intra-exam reproducibility. The measured PDFF values were similar for the two methods. Furthermore, measurement times further decreased in session 2 compared to session 1, suggesting that as researchers became familiar with the methods, they measured more quickly. Thus, measurement time could be further reduced with experience.

According to previous studies, the reproducibility and repeatability of liver fat measurements improve as the liver sampling area increases in size with a higher number of ROIs [22]. This might be due to the spatial heterogeneity of liver fat deposition because the segmental variation of fat deposition in the liver has been reported as approximately 12.5%–14.2% [29]. Furthermore, in previous studies, fat deposition was higher in the right lobe of the liver than in the left lobe, with or without a significant difference, whereas variability in fat deposition was higher in the left lobe than in the right lobe [29, 30]. However, a higher number of ROIs increases the time and labor intensity burden; hence, 4-ROI measurements have been suggested as a reasonable compromise between time burden and accuracy [21, 22, 30]. These simplified methods could be used in daily practice but might not be appropriate for clinical trials of new therapeutic drugs or treatment methods. Ideally, a method that samples from all of the imaged liver parenchyma should be used [31]. Whole-liver fat measurement has been attempted in previous studies [26, 27, 29, 31]; however, it has not been readily applied in clinical practice because of the lack of easily available software. A previous study showed a good correlation between Fat_vol and Fat_roi measured using a commercialized program, regardless of NAFLD severity [28]. These results are comparable with those of our study. In addition to agreement with these previous results, our study showed good agreement of Fat_vol values measured by two reviewers and demonstrated that using the Fat_vol method took significantly less time than the Fat_roi method. Furthermore, the range of the LOA was smaller for Fat_vol than for Fat_roi within each reviewer and between the two reviewers, suggesting that Fat_vol might be less affected by the heterogeneity of fat deposition in the liver than Fat_roi.

In a study by Campo et al. [22], measurement times decreased from 113–150 s to 69–82 s (approximately 28%–51% reduction) with a reduction in the number of ROIs from 12 to 4. The absolute measurement time was longer in our study than in their study, possibly because of differences in the definition of the start and end times. However, using Fat_vol rather than Fat_roi could achieve similar reduction (30%–63%) in the measurement time without sacrificing the measurement area.

Measurement times decreased in session 2 compared to session 1 for both reviewers and both methods (Fat_roi, 8%–14% reduction; Fat_vol, 7%–14% reduction). Furthermore, the measurement time was shorter for the more experienced reviewer (reviewer B) than for the less experienced reviewer (reviewer A). These results suggest that measurement times can be reduced with increasing experience. Considering that only two measurement sessions were performed in this study, we expect even more reduction as radiologists gain experience, and hope to confirm this in a future study.

This study had some limitations. First, most of the included patients had diffuse liver disease. However, because PDFF is usually evaluated in patients with suspected fat deposition, our results may be suitable for actual clinical situations. Second, only Fat_roi with 27 ROIs was compared with Fat_vol. Third, pathology was not used as a reference standard. As the accuracy of PDFF measured on MRI scans has already been established, this study focused on the inter- and intra-reproducibility and repeatability between methods and reviewers, as well as measurement time. Lastly, in patients with severe fatty liver disease, the heterogeneity of fat deposition in the liver increases, but this study did not include patients with severe fatty liver disease.

In conclusion, the measurement time of PDFF could be reduced by 30%–63% using Fat_vol compared to Fat_roi, while maintaining the consistency of the measured values. Therefore, compared to the conventional Fat_roi method, the Fat_vol method is expected to improve the workflow while maintaining measurement accuracy.

The online-only Data Supplement is available with this article at https://doi.org/10.13104/imri.2023.0020.

Supplementary Fig. 1

The correlation of PDFF between Fat_vol and Fat_roi for reviewer A (A) and reviewer B (B). Fat_vol, volumetric liver fat measurement method; Fat_roi, 27-regions of interest drawing method; PDFF, proton density fat fraction.