Development and validation of multivariable quantitative ultrasound for diagnosing hepatic steatosis

This study developed and validated multivariable quantitative ultrasound (QUS) model for diagnosing hepatic steatosis. Retrospective secondary analysis of prospectively collected QUS data was performed. Participants underwent QUS examinations and magnetic resonance imaging proton density fat fraction (MRI-PDFF; reference standard). A multivariable regression model for estimating hepatic fat fraction was determined using two QUS parameters from one tertiary hospital (development set). Correlation between QUS-derived estimated fat fraction(USFF) and MRI-PDFF and diagnostic performance of USFF for hepatic steatosis (MRI-PDFF ≥ 5%) were assessed, and validated in an independent data set from the other health screening center(validation set). Development set included 173 participants with suspected NAFLD with 126 (72.8%) having hepatic steatosis; and validation set included 452 health screening participants with 237 (52.4%) having hepatic steatosis. USFF was correlated with MRI-PDFF (Pearson r = 0.799 and 0.824; development and validation set). The model demonstrated high diagnostic performance, with areas under the receiver operating characteristic curves of 0.943 and 0.924 for development and validation set, respectively. Using cutoff of 6.0% from development set, USFF showed sensitivity, specificity, positive predictive value, and negative predictive value of 87.8%, 78.6%, 81.9%, and 85.4% for diagnosing hepatic steatosis in validation set. In conclusion, multivariable QUS parameters-derived estimated fat fraction showed high diagnostic performance for detecting hepatic steatosis.

www.nature.com/scientificreports/However, as qualitative assessment of B mode US echogenecity is likely dependent on the manufacturer, transducer, frequency, and experience of the operator, it can be negatively affected by interobserver variability 8,9 .
In recent years, various pulse echo quantitative ultrasound (PE-QUS) biomarkers, including ultrasound attenuation, backscatter coefficient, and speed of sound, have been proposed as objective tools with good reproducibility for hepatic steatosis quantification [8][9][10][11][12][13][14][15][16] .More recently, a few studies proposed US-derived fat fraction that combines attenuation and backscatter quantification using various mathematical approaches such as multivariable regression model or convolutional neural networks (CNNs) [17][18][19][20] , which presented measurement results in percentage, and showed good correlation with MRI-PDFF.This development is crucial because the quantitative presentation in percentage enhances the understanding of both patients and their physicians regarding the meaning of liver fat percentage and the goal of reducing this measurement 1 .However, as previous studies have shown that B-mode US has low sensitivity in diagnosing hepatic steatosis, especially for mild steatosis 21 , further validation studies including people without steatosis or with mild hepatic steatosis, are required to establish their clinical performance in detecting and staging MASLD as a screening or diagnostic tool.
Therefore, this study aims to develop and validate a multivariable QUS parameters-derived estimated fat fraction for diagnosing hepatic steatosis in patients with MASLD, using prospectively collected QUS parameters from a tertiary hospital and a health screening center, with the MRI-PDFF as the reference standard.

Methods
This was a retrospective cross-sectional study consisting of a secondary analysis of prospectively collected data from one tertiary hospitals and the other health screening center (

Study participants. Development set.
Between July 2020 and June 2021, participants who were referred to the radiology department for liver US for known or clinically suspected MASLD or were scheduled to undergo hepatectomy for liver donation were prospectively enrolled in the study performed at SNUH.The enrolled participants were aged 18 years and above.The exclusion criteria of the study include (a) evidence of liver disease other than MASLD; (b) substantial alcohol consumption; (c) long-term use of hepatotoxic or steatogenic medication; (d) previous liver surgery; and (e) contraindication for MRI 18 (Fig. 1).
Validation set.Between July 2022 and November 2022, participants who were 19 years or older and who underwent liver US for health screening were prospectively enrolled in the study performed at SMC.Among these participants, those who met the following exclusion criteria were excluded, which were: (a) evidence of chronic liver disease, including hepatitis B virus or hepatitis C virus-related diseases, (b) history of treatment for hepatic malignancy, (c) previous liver surgery, and (d) contraindication for MRI (Fig. 1).

QUS examination.
All US examinations were conducted by board-certified radiologists at each center (S.K.J., J.M.L., for SNUH and non-author radiologist for SMC with more than 9 years of experience, respectively).All participants underwent liver US examination using RS85 system (Samsung Medison, Seoul, Korea) using a convex probe (CA 1-7).All participants fasted at least 4 h before the US examination.After routine B-mode US examination of the liver parenchyma, QUS examination was performed in the right liver through the intercostal space.Participants were positioned in the supine position with the right arm at maximum abduction.For the measurement of two QUS parameters (tissue attenuation imaging [TAI] and tissue scatter-distribution imaging [TSI]), the operator placed a 2 × 3 cm fan-shaped region of interest (ROI) on the liver parenchyma, avoiding reverberation artifacts or large vessels.Areas with errors in parameter calculations, such as vascular structures, were automatically excluded from the calculation and presented as a vacancy on the TAI and TSI map.Operators obtained five measurements of QUS parameters (TAI and TSI) in each participant, which had a reliability index (R 2 value) of 0.6 or larger according to the vendor's recommendations.For each measurement, the five measurements were averaged as the representative value for each participant.
Chemical shift-encoded MRI PDFF.All participants underwent chemical shift-encoded liver MRI by using one of two 3.0-T systems (MAGNETOM Skyra [Siemens Healthineers, Erlangen, Germany] for development set; and Ingenia CX [Philips Healthcare, Amsterdam, Netherlands] for validation set) on the day of US examination whenever possible, or within a 14-day period.Detailed MRI protocol and parameters are summarized in Supplementary Material 1. Blinded to QUS results, one person at each institution (S.K.J., for the development set; and non-author radiologist for the validation set) placed a 1-cm circular ROI on each of the nine Couinaud segments.MRI-PDFF values from liver segments 5-8 were averaged and used as the reference standard for hepatic fat content.Mild, moderate, and severe hepatic steatosis were defined using MRI thresholds of 5%, 15%, and 25%, respectively 22 .

Statistical analysis.
Demographic and imaging data of patients were compared between the development and validation sets using t-tests for continuous variables and Chi-square tests for categorical variables.Using the development data set, we developed a multivariable regression model using two QUS parameters, which provided an estimated fat fraction as percentage based on the generalized linear regression to predict PDFF.The correlation between QUS-derived estimated fat fraction (USFF) and MRI-PDFF was evaluated using the Pearson correlation coefficient.The agreement between USFF and MRI-PDFF was assessed using a Bland-Altman analysis with 95% limits of agreement (LOA).Additionally, to investigate factors associated with poor agreement between USFF and MRI-PDFF, we compared demographic and imaging data of patients between cases with good agreement (within 95% LOA) and cases with poor agreement (exceeding 95% LOA) using t-test (Supplementary Material 2).Receiver operating characteristic (ROC) curve analysis was used to evaluate the performance of USFF for detecting any hepatic steatosis (MRI-PDFF ≥ 5%), moderate to severe hepatic steatosis (MRI-PDFF ≥ 15%), and severe hepatic steatosis (MRI-PDFF ≥ 25%).Pairwise comparisons of the areas under the ROC curve (AUCs) between USFF and TAI or USFF and TSI were performed using Delong test.In this pairwise comparison analysis, a Bonferroni-adjusted P values of less than 0.017 (0.05/3) was considered to indicate statistical significance.Cutoff values were determined using the maximal Youden index 23 , and corresponding performance parameters, including sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated in the development set.Additionally, cutoff values of USFF for sensitivity and specificity exceeding 95% were also calculated in the development set.For the assessment of performance of all given cutoffs, the sensitivity, specificity, PPV, and NPV values were computed in the validation set.Additional analysis was conducted within a subgroup after performing propensity score matching for sex and age (n = 133 each for development and validation set each) for evaluating the performance of USFF in assessing hepatic steatosis (Supplementary Table 2).All statistical analyses were performed using MedCalc version 18.11.6(MedCalc software, Ostend, Belgium).P < 0.05 was considered to indicate statistical significance.

Diagnostic performance of multivariable fat fraction estimator for hepatic steatosis in development and validation set.
The performance of USFF for diagnosing hepatic steatosis in the development set and validation set are summarized in Table 4.

Discussion
Our study demonstrated that estimated fat fraction from multivariable regression model using QUS parameters (USFF) was well correlated with MRI-PDFF (Pearson r = 0.799 and 0.824; for development and validation set, respectively), and the performance for diagnosing hepatic steatosis MRI-PDFF ≥ 5% was good in both development and validation set (AUCs of 0.943 and 0.924, for development and validation set, respectively).The good correlation between USFF and MRI-PDFF in a large number of study patients might have high clinical value and support clinical adoption of a less expensive modality with wider accessibility 24 .When applying a cutoff of 6.0%, which was derived from a development set (clinically suspected MASLD patients) for detection sensitivity of greater than 95% for MRI-PDFF ≥ 5% (rule-out criteria), USFF showed sensitivity, specificity, PPV, and NPV of 87.8%, 78.6%, 81.9%, and 85.4% for detecting hepatic steatosis in the validation set (health screening population).Although the previous study of US-derived fat fraction from other US vendors 17,24 also showed similar diagnostic performance, the distinct significance of our study arise from its incorporation of screening population, including many patients without steatosis or with mild steatosis.. Based on our study results, as the USFF can provide good diagnostic accuracy for detecting and grading hepatic steatosis.We believe that the USFF technique can be used as a useful adjunct to B-mode ultrasound examination for screening hepatic steatosis.
Of interesting note, although the populations of our study (development and validation set) had different distributions in the prevalence of hepatic steatosis, the USFF showed similar diagnostic performance in both populations.Therefore, we believe that USFF can be used as a tool for diagnosing and screening hepatic steatosis.In addition, since our validation set included health screening population, it included both non-alcoholic and alcoholic fatty liver disease.Our results demonstrated that USFF can be a promising screening tool for patients with hepatic steatosis from various etiologies, not only MASLD.Furthermore, we calculated different cutoff levels (rule-out and rule-in criteria) for wide clinical applications.If used as a screening tool, a cutoff value that Table 3. Diagnostic performance of QUS parameters and multivariable QUS-derived estimated fat fraction (USFF) for diagnosing hepatic steatosis.QUS quantitative ultrasound, AUC area under the receiver operating characteristic curve, 95% CI 95% confidence interval, MRI-PDFF magnetic resonance imaging proton density fat fraction, TAI tissue attenuation imaging, TSI tissue scatter-distribution imaging.*Receiver operating characteristic curve analysis was used to compute AUCs of USFF, TAI, and TSI in the evaluation of hepatic steatosis.A comparison of AUCs between USFF and TAI or USFF and TSI was conducted using Delong test, and a Bonferroni-adjusted P value of < 0.017 was considered to indicate statistical significance.In addition, the USFF is expressed as a percentage, analogous to MRI-PDFF 17,24 .This presentation of USFF is helpful in improving communication with patients and clinicians and might be beneficial for longitudinal therapy monitoring.Previous studies of convolutional neural network using QUS maps and ultrasound radiofrequency data also demonstrated excellent study results in assessing hepatic steatosis 18,20 .However, their reliance on extensive computing resources and databases poses significant challenges for practical implementation in clinical settings.Therefore, considering the ease of development and clinical application, statistical methods such as multivariable regression model can be preferred in such contexts.In line with this, efforts are being made to develop statistical models that propose US-based fat fraction in clinical vendors.Compared with those studies, which required deep learning models and RF US images, our study used multiple regression model using TAI and TSI measurement results and resulted in more straightforward and rapid calculations.Considering US as the most used imaging test for evaluating diffuse liver diseases and is also inexpensive, easily accessible imaging test with QUS measurements available in few seconds, the USFF seems to be promising for being used as a tool for screening hepatic steatosis.
In terms of grading of hepatic steatosis, among TAI, TSI, and USFF, USFF showed the highest AUC for assessing MRI-PDFF ≥ 5%, ≥ 15%, and ≥ 25% in both development and validation sets.We found that the diagnostic performance of all three parameters in assessing MRI-PDFF ≥ 25% was worse than MRI-PDFF ≥ 5% or ≥ 15%.These results are similar to previous studies 18,25 , and can be explained by the limitation of US signals which are weak in severe hepatic steatosis since it is already attenuated in the nearfield 20 .Therefore, we believe that QUSderived estimated fat fraction is suitable as screening test or diagnostic test for mild or moderate degree steatosis.
Our study has several limitations.First, although both the development and validation set were originally obtained from prospective data, the patient characteristics of both groups were significantly different.However, the validation group included a large number of patients without hepatic steatosis; this might be even more valuable.Second, although the sample size was large in both groups, all patients were from the same ethnic group, and therefore, the study is limited in terms of demographic and clinical variables.Third, additional studies are needed to validate the diagnostic performance of USFF against liver biopsy.Fourth, in this study, we evaluated a single vendor's ultrasound system.Further research using ultrasound system from various vendors is needed to validate our study results.Fifth, our study could not evaluate the liver stiffness value (MR elastography or shear-wave elastography data) of patients.To achieve a more comprehensive understanding of the disease status Table 4. Diagnostic performance of multivariable QUS-derived estimated fat fraction (USFF) for diagnosing hepatic steatosis in development and validation set.Unless otherwise noted, data are percentages, with numerators and denominators and 95% confidence intervals in parentheses.PPV positive predictive value; NPV negative predictive value; MRI-PDFF magnetic resonance imaging proton density fat fraction.www.nature.com/scientificreports/ in steatotic liver disease, it would be valuable to conduct additional validation studies that encompass both USFF and liver stiffness measurements.Finally, as our study did not include repeated measurements of USFF with an interval, we could not evaluate the capability of USFF for monitoring therapeutic effect of lifestyle changes or drug interventions.
In conclusion, multivariable QUS parameter-derived estimated fat fraction showed a good correlation with MRI-PDFF and high diagnostic performance for detecting hepatic steatosis in both development set of patients with suspected MASLD and validation data set of health screening population.Therefore, it can potentially be used as a screening imaging test to evaluate hepatic steatosis.
IRB No. for the prospective data collection: 2002-020-1099 for Seoul National University Hospital [SNUH]; and 2022-05-054 for Samsung Medical Center [SMC]).This retrospective study was approved by the Institutional review board of those two hospitals (SNUH [IRB No. 2209-142-1362] and SMC [IRB No. 2023-01-022]), and performed in accordance with the Declaration of Helsinki.Written informed consent was waived for this retrospective analysis because of the nature of the study.Two different study samples were used for developing and validating a statistical model to calculate US-based fat fraction from attenuation coefficient and backscatter coefficient, using MRI-PDFF as a reference standard.: a development study sample from SNUH (development set) in suspected MASLD population and a validation study sample from SMC (validation set) in health screening population.

Figure 2 .
Figure 2. Multivariable QUS parameters-derived estimated fat fraction versus MRI-PDFF scatter plot, along with the identity line and linear regression line for development set (a) and validation set (b). MRI-PDFF magnetic resonance imaging proton density fat fraction, CI confidence interval.

Table 1 .
Participant characteristics.Values are presented as mean ± standard deviation (range) or number (%) as appropriate.NAFLD nonalcoholic fatty liver disease, BMI body mass index, MRI-PDFF MRI proton density fat fraction.*P values were calculated using a t test for continuous variables or Chi-square test for categorical variables.

Table 2 .
Univariable and multivariable regression analysis for estimating hepatic fat fraction in development set.Values in parentheses are 95% confidence intervals.TAI tissue attenuation imaging, TSI tissue scatterdistribution imaging.*Multivariable generalized linear regression model was developed using TAI and TSI, which provided an estimated fat fraction percentage, using magnetic resonance imaging proton density fat fraction as the reference standard.

Development set P value* Validation set P value* AUC (95% CI) AUC (95% CI)
can improve sensitivity at the expense of specificity can be helpful.A cutoff value with optimal specificity can be helpful as a diagnostic tool.Using a dual cutoff level, USFF yielded PPVs of 98.1% and 93.1% for ruling in