Print
facebook
twitter
Linkedin
Introduction
Noninvasive assessment of liver fibrosis is a critical component of managing patients with chronic diffuse liver diseases, which affect millions of individuals worldwide [1]. Two-dimensional shear wave elastography (2D-SWE) provides real-time stiffness values (in kilopascals, kPa) for a selected region of the liver, offering a valuable tool for diagnosing significant fibrosis or cirrhosis without the need for biopsy [2–9]. However, like other ultrasound-based methods, 2D-SWE measurements can be influenced by technical factors and operator expertise, potentially leading to variability [10–12]. In addition, patient-related factors (such as obesity, a thick abdominal wall, a subcapsular measurement location, and a ratio of the region-of-interest [ROI] stiffness standard deviation [SD] to shear wave speed exceeding 0.15) have all been associated with increased measurement variability [12,13].
In conventional liver 2D-SWE, the operator manually selects an appropriate liver region with the best acoustic window (approximately 15–20 mm below the liver capsule), initiates image acquisition during a neutral breath-hold, and places an ROI to measure liver stiffness (LS) [7,8]. This process introduces operator-dependent variability in terms of acquisition timing, elastogram quality, and ROI placement [11,14,15]. Although approaches such as multiple measurements, confidence maps, median values, and quality metrics (e.g., interquartile range [IQR] or reliability indices) are recommended to increase reliability, consistently defining stable SWE images remains challenging [1,2,4,16–18]. This operator dependency has driven efforts to develop automated or semi-automated approaches, standardized protocols, and quality control measures, including reliability indices [1,16,19].
An automated algorithm can standardize both frame selection and ROI placement, thereby minimizing subjective judgment [19]. The fully automated 2D-SWE algorithm (EzSWI, Samsung Medison, Seoul, Korea) integrates an automated LS measurement system that uses reliability indicators calculated from the reliability measurement index (RMI) and elastogram uniformity [20]. The system also incorporates automated ROI placement, aiming to standardize measurements and reduce operator-dependent variability.
Although automation in elastography offers theoretical advantages, its effects on workflow efficiency and measurement reproducibility have not been systematically evaluated in clinical practice. It was hypothesized that automating the 2D-SWE workflow would reduce operator dependence by standardizing frame selection and ROI placement, while leaving the underlying physical principles of shear wave generation, propagation, and reconstruction unchanged. Improved repeatability and shorter examination times were therefore expected without compromising validity. Accordingly, this prospective study evaluated a fully automated 2D-SWE system in patients with diffuse liver disease, comparing it with semi-automated and manual methods in terms of examination time, observer variability, and measurement consistency.
Materials and Methods
Compliance with Ethical Standards
This study was approved by the Institutional Review Board of Seoul National University Hospital (IRB No. 2406-163-1551). All participants provided written informed consent. Samsung Medison (Seoul, Korea) provided financial support; however, the authors maintained full control over the data, analysis, and manuscript content.
Study Hypotheses and Endpoints
It was hypothesized that the automated workflow would reduce examination time and improve repeatability relative to the semi-automated and manual workflows by minimizing human measurement error. The primary endpoint was examination time for each workflow. Secondary endpoints included within-operator repeatability and between-operator reproducibility of LS measurements. The exploratory endpoint was between-system agreement (convergent validity) in a subset of participants who underwent repeat scanning on an independent vendor system (Canon Medical Systems, Otawara, Japan).
Participants
This prospective single-center study recruited participants scheduled for abdominal ultrasound at an academic tertiary care center between October 2024 and April 2025. The inclusion criteria were as follows: (1) adult participants (≥18 years) and (2) referral from a physician for diffuse liver disease, including chronic viral hepatitis, liver cirrhosis, and various causes of hepatic fibrosis such as hepatic steatotic disease. The exclusion criteria were as follows: (1) presence of solid focal liver lesions (≥1 cm), as such lesions can create local stiffness heterogeneity that may affect ROI placement consistency and measurement reproducibility; (2) acute hepatic disease; (3) a history of liver surgery or radiation therapy; and (4) refusal to participate in the study. The flow diagram of the study population is shown in Fig. 1.
Sample size justification
The sample size was determined based on two endpoints: (1) differences in examination time and (2) measurement reproducibility. For time comparison, approximately 28 participants were sufficient to detect a 10-second difference (α=0.05, power=90%, Cohen f=0.25, within-subject SD=7 seconds). A total of 40 participants were included, exceeding the minimum requirement. For reproducibility, 15 participants were sufficient to detect an intraclass correlation coefficient (ICC) increase from 0.80 to 0.95 (80% power, α=0.05). For intra-observer variability, 20 participants were sufficient to detect a 0.5 kPa mean difference (80% power, SD=1.0 kPa).
2D-SWE Acquisition
In all participants, LS measurements were performed using a Samsung Medison ultrasound system equipped with customized 2D-SWE capability and a CA1-7S convex broadband probe. All evaluations were conducted in the fasting state (>4 hours), with participants in the supine position and the right arm extended to optimize intercostal access to the right liver lobe. The 2×2.5 cm 2D-SWE frame box was placed within homogeneous parenchyma approximately 2 cm below the liver capsule and within 10 cm of the probe surface under a neutral breath-hold, in accordance with the World Federation for Ultrasound in Medicine and Biology guideline on liver ultrasound elastography [7]. LS values were then obtained by placing a 1-cm circular ROI. All machine settings were kept constant across measurement methods for each participant.
For evaluation of cross-platform agreement, all participants underwent additional measurements using a Canon Aplio i800 system with a convex probe (PVI-475BX) during the same session. Measurements on the Canon system were also performed using a standardized protocol, with participants in the supine position and using a right intercostal approach. During data acquisition, participants held their breath for several seconds while the convex probe was kept still. When a 2×2 cm frame box was positioned on the liver parenchyma approximately 2 cm below the liver capsule, the system automatically displayed a twin view of B-mode images and shear wave propagation maps. The propagation map was used to enhance measurement reliability, as smooth and parallel lines on the map indicated accurate and consistent stiffness evaluation. LS values were then obtained by placing a 1-cm circular ROI on the propagation map [4]. Median values from five valid measurements were recorded in accordance with device recommendations [21].
Measurement Methods
Three different SWE measurement approaches were evaluated (Fig. 2). All participants underwent measurements during a single session and in the same anatomical region. A standardized 2–3-minute interval was applied between each method to allow for system configuration. No clinical interventions were performed between measurements. Valid measurements required a reliability coefficient (R2) ≥0.8. Technical failure was defined as the inability to obtain five valid measurements. Operators were blinded to previous results through system design or by concealment of numeric values.
Method 1: Fully automated SWE (auto-acquisition+auto-ROI)
The fully automated 2D-SWE algorithm (EzSWI; Samsung Medison) handled both data acquisition and ROI placement. After initiating the 2D-SWE mode, the system continuously acquired elastogram frames along with a reliability indicator, calculated as the average of the RMI and elastogram uniformity [20]. Elastogram uniformity reflects the homogeneity of elasticity values and is calculated as the overall average of local SDs. As the elastogram becomes less uniform, the RMI value is proportionally down-weighted. Assuming uniform elasticity values across the image, the reliability indicator is assigned on a five-point scale: 5 for RMI exceeding 0.8; 4 for RMI between 0.6 and 0.8; 3 for RMI between 0.4 and 0.6; 2 for RMI between 0.3 and 0.4; and 1 for RMI of 0.3 or less. The reliability indicator scored each acquired frame and automatically selected those with a score of 3 or higher as acceptable quality, while excluding frames scoring 1 or 2. Technical failure was defined as the inability to obtain five valid measurements with reliability scores ≥3.
A complementary “Auto ROI Recommendation” algorithm then calculated the optimal circular ROI location by assessing local elasticity uniformity and the average local RMI within the measurement box, thereby mimicking expert user placement. Based on these analyses, the system automatically selected five acceptable-quality frames from the acquisition and placed a circular ROI in the most uniform, high-RMI region of each. LS values (kPa) were then calculated for each ROI without subjective input for frame selection or ROI positioning (Video clip 1).
Method 2: Semi-automated (auto-acquisition+manual ROI)
In this approach, the system assisted with acquisition by automatically flagging frames that met RMI and uniformity criteria (reliability indicator score ≥3). Operators then manually placed ROIs on the system-selected frames, avoiding areas of poor reliability indicated by the RMI maps.
Method 3: Manual (manual acquisition + manual ROI, RMI-guided)
Operators manually initiated and froze SWE acquisition based on visual assessment of elastogram quality and RMI maps, avoiding frames with large artifacts or inadequate shear wave propagation. After freezing acceptable elastograms, operators manually placed ROIs in representative, artifact-free regions.
Reproducibility Protocol
For intra-observer variability assessment, a radiologist with over 20 years of abdominal ultrasound experience performed methods 1 and 3 twice in succession on 21 participants. The radiologist was blinded to the numeric results of the first measurement while performing the second. For inter-observer variability, the remaining 15 participants were examined by two radiologists (the senior radiologist and another radiologist with 13 years of liver imaging experience), each independently performing methods 1, 2, and 3. Both radiologists were blinded to each other’s results, and participants were allowed a brief rest of a few minutes between scans performed by different operators.
Data Recording
Total examination time was recorded from the initiation of SWE acquisition to the point of obtaining final LS values. This included acquisition time (from SWE mode initiation to final frame capture) and measurement time (frame selection, ROI placement, and value recording). Research assistants operated digital stopwatch timers during examinations but did not perform measurements. To minimize bias, the operators were not informed of timing details.
Statistical Analysis
Continuous variables are presented as mean±SD or median with IQR. The Shapiro-Wilk test was used to assess data normality. Repeated-measures analysis of variance with Bonferroni-corrected post hoc tests was used to compare time measurements and LS values across the three methods. Measurement precision was evaluated using the coefficient of variation and IQR/median ratios, which were compared using the Friedman test. Within-session variability was assessed using the SD of five measurements per participant.
ICCs of mean LS values from each set of five measurements were calculated using a two-way model to assess intra-observer, inter-observer, and inter-method agreement. ICCs >0.90 indicated excellent reliability. Bland–Altman analysis was performed to assess agreement between methods and observers, with limits of agreement defined as the mean difference±1.96 SD. Comparisons with the Canon i800 ultrasound system were performed using paired t-tests and ICC analysis to assess cross-platform agreement. Statistical analyses were performed using MedCalc version 20.0 (MedCalc Software, Ostend, Belgium), with P-values of less than 0.05 considered to indicate statistical significance.
Results
Participant Characteristics
Overall, 40 participants were recruited for this study. No incidental focal liver lesions were detected during ultrasound examination; therefore, no participants were excluded after enrollment. The cohort comprised 23 men and 17 women (mean age, 55 years; range, 20 to 83 years). All 40 participants were included in both the inter-method and cross-platform measurements. For the subgroup analyses, 21 and 15 participants were allocated to the intra-observer and inter-observer measurement sessions, respectively. As these numbers satisfied the predefined sample size calculations, the remaining four participants were not included in either subgroup measurement. As anticipated during sample size planning, two technical failures occurred: one participant (body mass index [BMI], 33.71 kg/m2) was excluded from the cross-platform analysis due to the inability to acquire usable data with the Canon ultrasound system, and another participant (BMI, 39.02 kg/m2) was excluded from the intra-observer agreement analysis because a second measurement could not be obtained with the Samsung Medison ultrasound system. Consequently, 39 participants were included in the cross-platform analysis, 20 in the intra-observer agreement analysis, and 15 in the inter-observer agreement analysis, all meeting the minimum sample size requirements defined during study planning (Fig. 1). The mean BMI was 26.5±4.5 kg/m2. The underlying etiologies of liver disease included chronic hepatitis B (n=27), chronic hepatitis C (n=2), alcoholic or nonalcoholic steatohepatitis (n=8), and cryptogenic or other causes (n=3) (Table 1). Five participants were clinically classified as having cirrhosis, with LS values ranging from approximately 7.8 kPa to 41.9 kPa. RMI-based scoring during automatic acquisition yielded a mean frame quality score of 4.2 (out of 5), indicating generally good elastogram quality across the study population.
LS Measurements
Mean LS values were comparable across all three methods: fully automated (6.52±6.16 kPa), semi-automated (6.49±6.02 kPa), and manual (6.60±6.01 kPa), with no statistically significant differences among these (P=0.556). Pairwise mean differences were minimal: automated vs. semi-automated, −0.03±0.52 kPa (P=0.730); automated vs. manual, 0.08±0.90 kPa (P=0.556); and semi-automated vs. manual, 0.11±0.93 kPa (P=0.451). Fig. 3 illustrates the within-session variability for each method, shown as the distribution of the SD of five measurements per participant. Although the fully automated method (median SD, 0.16) showed a trend toward lower variability compared with the semi-automated (median SD, 0.195) and manual (median SD, 0.19) methods, the difference was not statistically significant (P=0.456).
The IQR/median ratio was also lower for the fully automated method (7.23%±5.54%) than for the semi-automated (7.66%±4.18%) and manual (8.63%±5.52%) methods, although the difference was not statistically significant (P>0.05). Agreement across the three methods was exceptionally high (ICC, 0.997; 95% confidence interval [CI], 0.995 to 0.998).
Examination Time
The fully automated method significantly reduced total examination time (14.38±5.01 seconds) compared to the semi-automated (30.46±7.11 seconds, P<0.001) and manual (33.05±7.10 seconds, P<0.001) methods. The difference in total time was also statistically significant between the semi-automated and manual methods (2.59±1.27 seconds, P=0.048) (Table 2, Fig. 4). Time savings were achieved primarily in the measurement phase (automated: 6.91±2.54 seconds, semi-automated: 22.98±6.24 seconds, manual: 24.73±6.59 seconds; P<0.001).
Intra-observer and Inter-observer Variability
Intra-observer agreement was excellent for both the fully automated (ICC, 0.971; 95% CI, 0.928 to 0.988) and manual (0.974; 95% CI, 0.934 to 0.989) methods. The mean difference between repeated measurements with the automated method (0.15±0.71 kPa) was smaller than that obtained with the manual method (0.24±0.95 kPa) (Table 3, Fig. 5). Inter-observer agreement was also excellent for the fully automated (0.984; 95% CI, 0.951 to 0.995), semi-automated (0.992; 95% CI, 0.977 to 0.997), and manual (0.975; 95% CI, 0.926 to 0.992) methods. The mean difference between radiologists using the automated method (0.14±2.24 kPa) was smaller than that using the manual method (0.45±2.62 kPa) (Table 3, Fig. 5).
Cross-Platform Comparison
Among the 39 participants who underwent comparative measurements, good agreement was observed between the study system’s automated measurements and those obtained using the Canon system (median, 6.58±6.23 kPa vs. 7.24±7.19 kPa; P=0.425), with a mean difference of 0.66±5.09 kPa. ICCs between the Canon system and all three methods were good (automated, 0.833; semi-automated, 0.861; manual, 0.837), indicating robust cross-platform agreement regardless of measurement method.
Discussion
This prospective study was designed to evaluate workflow efficiency and measurement reproducibility across automated, semi-automated, and manual 2D-SWE workflows, without assessing diagnostic accuracy. Because the principles of shear-wave excitation and velocity measurement are identical across all workflows, automation is not expected to improve diagnostic accuracy but rather to increase workflow efficiency and measurement consistency. The results clearly demonstrate that the fully automated 2D-SWE method, which incorporates automated frame selection and ROI placement algorithms, significantly improves clinical workflow and measurement reliability in participants with diffuse liver disease. Compared with the semi-automated and conventional manual methods, the fully automated approach substantially reduced total examination time (by approximately 16–19 seconds per participant) while maintaining high consistency and reproducibility of LS measurements. In the cross-platform analysis using Canon 2D-SWE in nearly all analyzable participants (n=39), LS measurements showed good agreement for all three workflows. These findings support the convergent validity of the automated workflow across different vendor platforms, although they do not establish its diagnostic accuracy.
The time savings achieved with the automated approach occurred primarily during the measurement phase, where automated frame selection and ROI placement eliminated operator-dependent variability. Moreover, all three methods yielded comparable LS values (automated, 6.52±6.16 kPa; semi-automated, 6.49±6.02 kPa; manual, 6.60±6.01 kPa; P>0.05), demonstrating that automation does not compromise measurement reproducibility. Although the per-patient time saved (16–19 seconds) may appear modest, it can accumulate in high-volume clinical settings and, importantly, reduce operator dependence by standardizing frame selection and ROI placement. For inexperienced operators or uncooperative participants, the automated system’s capacity to rapidly capture high-quality frames may determine whether an examination is successful [7,22]. To these authors’ knowledge, this is one of the first clinical studies to directly compare a fully automated SWE method with standard manual methods in the same participants, and the findings strongly support integrating such technology into routine liver elastography practice [23].
Minimizing operator dependence is essential for accurate and reproducible LS measurements [1]. Initially, it was hypothesized that the fully automated method would reduce measurement variability through two mechanisms: automated frame selection based on reliability indicators (elastogram uniformity and RMI) and standardized ROI placement in regions with optimal local elasticity uniformity and RMI scores. The results demonstrated excellent reproducibility across all methods, with the automated approach showing performance comparable to conventional techniques. Intra-observer ICC values were similar between the fully automated (0.971) and manual (0.974) methods, as were inter-observer ICC values (0.984 vs. 0.975, respectively). Although the fully automated method showed a numerically smaller mean difference between repeated measurements (0.15±0.71 kPa vs. 0.24±0.95 kPa for manual) and a lower IQR/median ratio (7.23%±5.54% vs. 8.63%±5.52%), these differences did not reach statistical significance. When LS differences between intra-observer and inter-observer measurements were plotted, the 95% limits of difference were 1.4 kPa for the fully automated method and 1.5 kPa for the manual method in intra-observer measurements, and 4.3 kPa and 5.1 kPa, respectively, in inter-observer measurements. Although the 95% limits of difference were narrower with the fully automated method than with the manual method, the acceptable margin for ensuring reproducibility has not been validated; therefore, caution is warranted when interpreting results, particularly in patients with low LS values, where such differences could alter clinical classification. These findings suggest that while automation provides consistent measurement precision, its primary advantage lies in workflow efficiency rather than a substantial improvement in reproducibility over skilled manual operation. Beyond workflow efficiency, the standardization achieved through automation has important implications for operator training and quality assurance across different experience levels. Therefore, the automated workflow is recommended as the default, with manual or semi-automated methods serving as backups for suboptimal acquisitions. Clinical expertise remains essential for proper image acquisition, patient positioning, and final diagnostic interpretation. The cross-platform comparison demonstrated good agreement between the study system’s automated measurements and those obtained using the Canon system (ICC, 0.833), with a mean difference of 0.66±5.09 kPa. This finding is particularly relevant for clinical practice, as it shows that the automated method yields measurements consistent with those of other established elastography platforms. The high ICC values between all three methods and the Canon system (ranging from 0.833 to 0.861) further suggest that the choice of measurement method does not significantly affect cross-platform agreement, although the fully automated method provides considerable time advantages.
Interestingly, the total time saving between the semi-automated and manual methods was only 2.59±1.27 seconds (P=0.048), whereas the time saving between the fully automated and semi-automated methods was 16.07±5.78 seconds (P<0.001). The primary difference between the fully automated and semi-automated methods was ROI selection, while the distinction between the semi-automated and manual methods lay in frame selection. This finding implies that most time can be saved through the automation of ROI selection rather than frame selection. ROI placement requires deliberate judgment, careful adjustment of location, and repeated fine-tuning to avoid large vessels or heterogeneous regions, making it inherently more time-consuming and dependent on the operator. By eliminating these subjective steps, fully automated ROI selection not only reduces examination time but also minimizes variability.
Although automation appears to reduce overall variability, its effectiveness in improving success rates or minimizing variability in patients who are difficult to measure—such as those with high BMI or advanced cirrhosis [12,24–27]—remains uncertain and warrants further investigation. In the present cohort, two participants with high BMI did not yield valid measurements using the manual method but were successfully measured with the fully automated method, suggesting a potential benefit. However, an outlier in Fig. 5D exhibited a substantial mean difference greater than 6 kPa, a finding also observed with the fully automated method in Fig. 5C. Thus, whether automation can reliably reduce LS variability in challenging patients remains to be determined.
Several limitations should be acknowledged. First, the sample size (40 participants) was modest. Although it was sufficient to demonstrate significant differences in examination time, a larger multicenter study would increase the generalizability of the findings. Second, all participants in this cohort were able to comply with breath-hold instructions. The benefits of automation may be even more pronounced in participants who cannot adequately hold their breath, highlighting an important direction for future studies focusing on technically challenging populations. Third, the absence of a gold standard precludes definitive statements about accuracy. The cross-platform analysis provides evidence of relative agreement only; future studies should directly compare automated 2D-SWE with a gold standard to confirm clinical validity and generalizability. Fourth, results were not stratified by fibrosis stage or LS value. Improvements in variability may be more significant in stiffer livers, for which manual ROI placement is more difficult. Fifth, the cohort primarily consisted of patients with chronic hepatitis B (67.5%), and performance variations of the automated method across different liver disease etiologies were not assessed. In particular, hepatic steatosis may affect elastogram quality, which limits the generalizability of the findings. In addition, the results are based on data obtained using a single ultrasound vendor’s equipment; therefore, the observed improvements in workflow and consistency may not be immediately applicable to other 2D-SWE platforms without further validation. Potential conflicts of interest should also be considered when interpreting the results, as the study was supported by the vendor, although all analyses and interpretations were conducted independently by the authors. Cross-platform studies are warranted to determine whether the benefits of automation are vendor-specific or represent a broader technological advancement. Finally, despite the clear benefits observed with the automated method, including reduced operator-dependent variability and shorter examination times, cost-effectiveness and real-world feasibility across diverse clinical environments were not assessed in this study and remain important areas for future investigation.
In conclusion, the fully automated 2D-SWE protocol markedly reduces scanning time, by approximately 55% compared to conventional methods. Moreover, by maintaining high reproducibility of LS measurements, automated elastography could facilitate the standardization of liver fibrosis assessment across diverse clinical settings, thereby promoting more consistent patient management and improving the quality of longitudinal monitoring. Future studies should explore the impact of this technology on diagnostic accuracy for fibrosis staging and its integration with other imaging modalities.
Download Citation
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
