Liver Fat Quantification With Ultrasound: Depth Dependence of Attenuation Coefficient

The primary aim was to estimate the influence of various depths on ultrasound attenuation coefficient (AC) of multiple vendors in the liver. The secondary aim was to evaluate the impact of region of interest (ROI) size on AC measurements in a subset of participants.

Algorithms for estimating the ultrasound (US) attenuation coefficient (AC), which is positively related to liver fat content, are currently commercially available, and the most used are based on AC. [3][4][5] US attenuation is defined as the cumulative loss of energy in a medium due to scattering and absorption as the acoustic wave travels through the medium.The amount of attenuation varies with tissue type and these changes in values allow attenuation to detect changes in tissue microstructure.Attenuation can be quantitatively estimated from the echo signal received in the transducer by accounting for both amplitude and frequency changes in the content of the signal thus facilitating local estimation of attenuation. 4mong different vendors, there is a lack of consensus on the best protocol for estimating the AC.In a recent study aimed at assessing factors that affect the intra-observer and inter-observer repeatability of AC imaging measurements using various positions of the transducer or size of the region of interest (ROI), a decrease in AC values by increasing the ROI's depth from 2 to 3 cm was noticed (Ferraioli et al, presented at the 2022 annual meeting of the Radiological Society of North America).In that study, the attenuation imaging (ATI) algorithm embedded in the Aplio iseries US systems (Canon Medical System, Japan) was used.
The primary aim of our study was to estimate the influence of various depths on estimating the AC of various vendors in the liver.The secondary aim was to evaluate the influence of ROI size on AC measurements in a subset of participants using one vendor algorithm.

Materials and Methods
This Institutional Review Board (IRB)-approved Health Insurance Portability and Accountability Act (HIPAA)-compliant retrospective study was performed in two centers (site 1 and site 2): the Ultrasound Unit of "Fondazione IRCCS Policlinico San Matteo," Department of Clinical, Surgical, Diagnostic and Pediatric Sciences, University of Pavia, Italy (site 1) and Southwoods Imaging Center, Youngstown, Ohio, USA (site 2).
In site 1, AC measurements were carried out with the Aplio i800 US system software version 6.5 using the i8CX1 convex transducer and the ATI (ATI-Canon) algorithm.Moreover, measurements were also performed with the ultrasound-derived fat fraction (UDFF), the algorithm combining AC and backscatter coefficient implemented in the Sequoia US system (Siemens Healthineers, Germany) software version VA40.For UDFF, the AC data were extracted offline from the raw data stored on a hard disk.For AC-Canon, the data of 63 individuals enrolled in previous published studies performed to assess the diagnostic accuracy of AC-Canon 6,7 were extracted from the database and re-analyzed for a different purpose if at least five images suitable for redoing the measurements at all the depths set in the present study had been recorded.The UDFF data were obtained on consecutive participants studied for a routine workup.
In site 2, AC measurements were performed with the Epiq7 US system (Philips Medical Systems, the Netherlands) using the C5-2 convex transducer and the ATI-Philips algorithm software version 9.0.5.Data of 30 consecutive individuals enrolled in a study performed to compare the Philips and Canon algorithms for ATI 8 were reviewed and 5 images with AC-Philips at each depth using 1 and 3 cm ROI were measured and recorded.An additional 30 patients were studied using a 3 cm ROI at each depth providing 60 patients assessed with the AC-Philips.
Examinations were performed by expert operators in both sites on participants fasting for at least 4 hours.The transducer was positioned in the intercostal space, and measurements were obtained on the best quality image, that is, the one with fewer vessels and the strongest B-mode signal without artifacts.
Canon and Philips algorithms give a color-coded image of the AC and automatically filter out structures such as vessels or ligaments while calculating the AC.The field of view and the ROI are useradjustable with both algorithms.With the ATI-Canon, the quality of the AC measurement is assessed by an R 2 value, a linear regression coefficient of determination.The font color of the displayed R 2 value and of the AC value represents the quality of the measurement: white for high quality (R 2 ≥ 0.90), yellow for moderate quality (R 2 from 0.70 to 0.89), and red for low quality (R 2 value below 0.70) (Figure 1A).With the ATI-Philips, the quality of the AC measurement is assessed by a confidence map which is an aggregate of multiple metrics and ranges from 100% (maximum confidence) to 0% (no confidence).A confidence threshold ≥60% indicates a high-quality measurement (Figure 1B).For this study, since the measurement quality inevitably degrades with depth, the quality threshold was set by the operators at a moderate level (R 2 ≥ 0.70 for AC-Canon and confidence map threshold ≥30% for AC-Philips).One operator at each site (GF and RGB) reviewed the images to perform multiple AC measurements at different depths with algorithms from several vendors, and measurement with different ROI sizes using one vendor algorithm.They had 37 and 32 years of experience in B-mode US.Measurements with AC-Canon and AC-Philips were taken by Figure 1.A, Attenuation coefficient obtained with the Canon algorithm at a depth of 2 cm (upper edge of the region of interest, which is the rectangle with white borders inside the large, color-coded field of view).The value, in decibel/centimeter/megahertz, is shown on the left bottom side of the image together with the R 2 value that is a quality factor and shows high quality of the acquisition.B, Attenuation coefficient obtained with the Philips algorithm at a depth of 2 cm (upper edge of the region of interest-dotted circle).The value, in decibel/centimeter/megahertz, is shown on the left bottom side of the image.The confidence map (rejection algorithm) is listed in the left upper image.In this case the confidence threshold (CT) is 60%.The algorithm evaluates the quality of the image and assigns a value between 0 and 100% with 0% meaning no confidence and 100% high confidence.This is used to remove artifacts from the image.C, Ultrasound derived fat fraction (UDFF) algorithm available on Siemens Sequoia ultrasound system.The UDFF value is given as percentage of fat and is shown on left upper side of the image.The depth shown below the UDFF value (4.6 cm) refers to the distance between the transducer and the center of the region of interest.The upper edge of the region of interest is positioned at 1.5 cm below the liver capsule as per the manufacturer recommendation.
positioning the ROI upper edge at 2, 3, 4, and 5 cm below the liver capsule.Care was taken not to include reverberation artifacts, that may arise near the liver capsule, or values with a low signal-to-noise ratio in the ROI.With the AC-Canon algorithm, the reverberation artifacts appear as a dark orange area whereas a low signal-to-noise ratio appears dark blue.With the AC-Philips algorithm, both artifacts are highlighted with red, meaning low quality, in the confidence map.The ROI's size was set at 3 cm.
For UDFF, the ROI has a fixed size (length of 40 mm, upper and lower width of 30 and 40 mm, respectively).The measurements were performed with the deep abdominal transducer (DAX: 1.0-3.5 MHz) (Figure 1C).For the Siemens system, both UDFF and AC measurements at different depths were obtained from the raw data.The measurements were obtained with the upper edge of the sample box at 1.5 cm below the liver capsule, the depth recommended by the manufacturer for UDFF measurements, and at 2 and 3 cm.Beyond this distance, they were unreliable in all cases because there was a very low signal-to-noise ratio.
A subgroup of 30 participants performed on the Philips Epiq7 was analyzed for the effect of ROI size.For this purpose, measurements performed at 2, 3, 4, and 5 cm depth with a 1 and 3 cm ROI were compared to assess the agreement.
The median value of 5 acquisitions was used for the analysis.

Statistical Analysis
Power of the study: According to the "rule of thumb" of 10 records per "candidate variable," 9 where "candidate variable" actually means number of parameters to be estimated in the model, we considered a minimum sample size of 50 participants adequate to build a multivariable model with a maximum of five covariates.Descriptive statistics were produced for the demographic characteristics of this study sample.Quantitative variables were expressed as the mean value and standard deviation (SD), and qualitative variables were summarized as counts and percentages.
Univariable and multivariable (including sex, age, stiffness, skin-to-liver capsule distance) linear regression models were fitted using AC value as the dependent variable and the depth at which measurements were taken as the independent variable.Body mass index and waist circumference were not entered into the regression models because of collinearity with the skin-to-liver capsule distance.Results are expressed as beta coefficients with their 95% CI and presented with term-specific P values.The coefficient represents the mean variation of outcomes for unit change of quantitative predictors or between levels of categorical or ordinal predictors.
Differences between AC values obtained with two different ROI's size, that is, 3 and 1 cm, in the subgroup of 30 participants assessed with the AC-Philips algorithm were evaluated with t-test for paired data.Lin's concordance correlation coefficient (CCC) was used to evaluate the agreement between values obtained with two different ROI's size, that is, 3 cm and 1 cm, in the subgroup of 30 participants assessed with the AC-Philips algorithm. 10CCC ranges in values from 0 to +1.The agreement was classified as poor (0-0.20),fair (0.21-0.40), moderate (0.41-0.60), good (0.61-0.80), and excellent (0.81-1.00). 11 < .05 was considered statistically significant.All tests were two-sided.ADS performed the data analysis with the STATA statistical package (release 17.0, 2021, Stata Corporation, College Station, Texas, USA).

Results
Sixty-three participants (34 females, mean age: 51 AE 14 years) were studied with the AC-Canon, 60 participants (46 females, mean age: 57 AE 11 years) with the AC-Philips, and 50 participants (25 females, mean age: 61 AE 13 years) with the AC-Siemens.The characteristics of the three study cohorts are reported in Table 1.Steatosis was assessed with B-mode US in the AC-Canon and AC-Siemens cohorts and was present in 32 (50.8%) and 18 (36.7%)participants, respectively.In the AC-Philips cohort, steatosis assessed with magnetic resonance imaging proton density fat fraction (PDFF) or hepato-renal index was present in 28 (46.7%)participants.

Effect of ROI's Depth
With all three software, there was a progressive decrease of AC values increasing the depth (Table 2 and Figure 2).Table 2 reports the AC values at different ROI's depths.
For AC-Canon, univariable regression analysis showed that the decrease was of 0.052 dB/cm/MHz (95% CI À0.062 to À0.042; P < .001)per 1 cm increase of the depth.This finding was confirmed with the multivariable model (coefficient À0.049; 95%CI À0.060 to À0.038 P < .001),and AC value was also significantly affected by the skin-to-liver capsule distance (P < .001)(Table 3).Comparable results were obtained with the AC-Philips (coefficient: À0.058 dB/cm/MHz at univariable and multivariable linear regression analysis) and the AC-Siemens (coefficient: À0.076 and À0.081 dB/cm/ MHz at univariable and multivariable linear regression analysis, respectively) (Tables 4 and 5).For the AC-Philips, the AC value was not affected by the skin-toliver capsule distance or any other variable.

Effect of ROI Size
In the subgroup of participants that were assessed with the AC-Philips algorithm using 1 cm ROI, measurements at 2 cm were not feasible because no pixels or only a few highly attenuating pixels were present in the ROI.As observed with the 3 cm ROI, there was a decrease in AC values with depth; moreover, the AC values with 1 cm ROI were significantly higher (P < .001)than those obtained with the 3 cm ROI at all depths (Table 6).
The agreement between AC values obtained with 1 and 3 cm ROI was excellent (CCC 0.82 [95% CI: 0.77-0.88]),indicating that the AC values followed the same trend even though they were not the same (Figure 3).

Discussion
The results of this study show that the AC values depend on the depth of the measurement and that there is a progressive decrease of the values that is directly related to the depth.This finding is of utmost relevance because thresholds for detecting and grading liver steatosis might vary depending on the ROI's depth for AC measurements.Indeed, several studies a Upper edge of the measurement box positioned at 1.5, 2, and 3 cm below the liver capsule.CI, confidence interval.that have assessed the performance of AC algorithms for fat quantification using PDFF or liver histology as reference standard have been published. 3,4,12They reported slightly different optimal cutoff values for the liver steatosis diagnosis even when the same vendor's AC algorithm was used.For instance, for the ATI-Canon, this value ranged from 0.59 to 0.69 dB/cm/MHz. 4These different results might be due to differences in the acquisition protocol.The results of our study suggest that ROI's depth is a major influencing factor that must be considered because it could be a source of bias.Fat quantification with AC algorithms is a new tool for which a standardized acquisition protocol is strongly needed; otherwise, its value in the diagnostic work-up of patients would be severely undermined.
In AC-Canon and AC-Siemens cohorts, the skinto-liver capsule distance affected the AC values, whereas this influence was not observed in the AC-Philips cohort.We hypothesize that the higher skinto-liver capsule distance in the AC-Philips cohort attenuated the influence on AC values at different depths.Interestingly, replacing the skin-to-liver capsule distance with body mass index in the univariable and multivariable models (result not shown), this finding did not change.Another hypothesis is that the rejection rate of the Philips software may work differently on shallow depths.
Of note, not only the depth but also the size of the ROI affected the AC value.Indeed, in a subset of our cohort, we obtained higher AC-Philips values with a smaller ROI.This finding is likely due to the fact that, with a size of 1 cm, the deep edge of the ROI is 2 cm closer to the liver capsule respect to a size of 3 cm.Since the AC is an average of the colorcoded values in the ROI, for each depth of the ROI, the mean sampling occurs at a deeper level for a larger ROI box than the mean sampling for a smaller ROI.In this setting, the effect of depth on AC values that was found in our study plays a major role.In this regard, it must be underscored that the differences observed between 1 and 3 cm ROI are in line with the coefficient of 0.058 dB/cm/MHz for AC-Philips that was obtained in the multivariable regression analysis.This finding is of utmost importance because the ROI size may be a confounding factor in studies assessing the repeatability or accuracy of the AC parameter for fat quantification or in the follow-up of patients.
Due to the variability of the AC with depth, a strict protocol defining the depth of the ROI from the liver capsule and its size must be defined for accurate correlation with the reference standard.
Studies in phantoms show that the lowest bias in AC measurement is obtained when the elevational focus matches the ROI depth. 13Most abdominal transducers have an elevational focus close to 6 cm.Hence, ROI close to elevational focus should likely be the default setting because AC is overestimated at depths lower than elevational focus and underestimated at depths higher than elevational focus.Indeed, as shown by an earlier investigation (Ferraioli et al, 2022 RSNA meeting) with the ATI-Canon algorithm, the highest repeatability of the AC measurement was obtained on images of the best quality, by positioning the upper edge of the ROI at 2 cm below the liver capsule, avoiding including reverberation artifacts, and with ROI size of 3 cm.Considering that the mean thickness of the subcutaneous tissue was 1.86 cm (0.25), the center of the ROI was close to 6 cm in the majority of cases.
The major strength of our study is that similar values of the coefficient of multivariate regression analysis of AC with depth were obtained in different cohorts using US systems from different manufacturers.Another strength is that the protocol for AC acquisition was the same in the two centers.
There are limitations to this study.First, measurements at different depths were performed on images recorded for other purposes.Hence, they were not available for some subjects due to poor quality in the far field of the previously recorded image.Second, due to the retrospective nature of the study, data on AC-Canon previously acquired in the North American center were not available anymore; therefore, it was not possible to compare the results obtained with the AC-Canon and AC-Philips in the same cohort.Third, we did not assess whether similar results were obtained with all algorithms based on ATI from different vendors.However, our personal experience with other manufacturers' software is in keeping with the present study's findings.Third, biochemical tests were available only for the AC-Canon cohort.Nonetheless, the analysis of the data in this cohort showed that there was not any influence of the biochemical tests on the results.
In conclusion, there is a depth and an ROI size dependence in measuring the AC obtained in the liver.This can substantially affect the results.A standardized acquisition protocol with a fixed depth and size of the ROI needs to be developed to minimize differences in AC measurements and to assess changes in serial measurements reliably.

Figure 2 .
Figure 2. The figure shows the decrease in attenuation coefficient values observed with the depth.Vertical axis: attenuation coefficient values; horizontal axis: distance of the region of interest from the liver capsule.A, Canon; B, Philips; C, Siemens.

Figure 3 .
Figure 3. Agreement between AC-Philips values obtained with a 3 cm ROI (vertical axis) and a 1 cm ROI (horizontal axis).

Table 1 .
Characteristics of the Three Study Cohorts AC, attenuation coefficient; SD, standard deviation; BMI, body mass index; AST, aspartate transaminase; ALT, alanine transaminase; kPa, kilopascal.aAssessed with B-mode ultrasound.bassessed with magnetic resonance imaging proton density fat fraction in 30 participants and with hepato-renal index in the other 30 participants.

Table 2 .
Mean Values of the Attenuation Coefficient Obtained at Different ROI's Depth aIt is referred to the distance from the liver capsule of the upper edge of the measurement box.

Table 3 .
Univariable and Multivariable Linear Regression Analysis of AC-Canon Values Obtained in 63 Cases With a Region of Interest of 3 cm Positioned at Different Depths Upper edge of the measurement box positioned at 2, 3, 4, and 5 cm below the liver capsule. a

Table 4 .
Univariable and Multivariable Linear Regression Analysis of AC-Philips Values Obtained in 60 Cases With a Region of Interest of 3 cm Positioned at Different Depth a Upper edge of the measurement box positioned at 2,3, 4, and 5 cm below the liver capsule.

Table 5 .
Univariable and Multivariable Linear Regression Analysis of AC-Siemens Values Obtained in 50 Cases With the Standard 4 cm Region of Interest Positioned at Different Depths

Table 6 .
AC-Philips Values Obtained in the Subset of 30 Participants at Different Depths With an ROI of 3 and 1 cm AC, attenuation coefficient; dB/cm/MHz, decibel/centimeter/megahertz; ROI, region of interest.a It is referred to the distance from the liver capsule of the upper edge of the measurement box.