Introduction

The amount, distribution, and quality of fat in the core region of the body have significant clinical implications1,2. Visceral adiposity has been associated more accurately with cardiometabolic abnormalities than standard body measurements such as body mass index (BMI)3. Both the area and radiation attenuation of visceral fat have been associated with important clinical outcomes including mortality2. In the Framingham Heart Study, decreasing abdominal visceral fat attenuation was associated with worse cardiovascular outcome and greater cardiometabolic risk2, 4, while increased pericardial and intrathoracic fat was associated with decreased adiponectin and increased metabolic risk5. In patients with liver disease, higher visceral fat attenuation has been associated with higher mortality in patients with chronic liver disease and hepatocellular carcinoma6,7,8.

Visceral fat measurements can be inferred from methodologies such as anthropometry, dual energy X-ray absorptiometry (DXA), and bioelectrical impedance analysis (BIA). Higher-cost methods such as magnetic resonance (MR) and computed tomography (CT) can definitively delineate visceral from subcutaneous fat, however, MR is more expensive, slower, and less common than CT. Therefore, using existing CT scans acquired for clinical indications can be a convenient source of detailed body composition information without additional cost or risk of radiation. However, routine CTs have differential protocols, including differences in contrast administration, and different ranges of Hounsfield Unit (HU) values have been used to define fat pixels, both of which may affect fat measurements.

A few studies have investigated the effect of contrast phase on CT visceral fat measurement in clinical cohorts, generally reporting that VFA is higher in non-contrast compared contrast phase scans, though effect sizes varied widely and different fat HU ranges were used9,10,11,12,13,14.

The definition of visceral fat measured from CT depends upon the HU range used to define and extract fat pixels from imaging. Commonly used HU ranges include − 150 to − 50 HU9,10,11, − 195 to − 45 HU15,16,17, − 190 to − 30 HU18,19,20,21, and − 250 to − 50 HU22. We used − 205 to − 51 HU based on the adult fat threshold preset in Materialize Mimics software (version 17.0).

Studies recommend measuring visceral fat at the L4-L5 level to maximize correlation between axial fat areas and whole body fat volumes18, 19, 21, 23. However, the L4-L5 level is not available in all clinical CT scan protocols.

Previous work has reported healthy reference values for skeletal muscle measures using similarly large cohorts of kidney donor candidates24,25,26.

To our knowledge, no study has reported sex-specific, contrast and non-contrast reference values for visceral fat area and radiation attenuation in a large, healthy, US adult cohort, utilizing different fat HU ranges. The aim of this study is to quantify the effect of contrast and fat HU range on fat measurements and report reference values. Understanding these effects on visceral fat area and density would allow us to better leverage the important information that is present in CT scans performed for clinical indications.

Results

Cohort summary

The majority of subjects (1093, 65.2%) eventually donated a kidney while the remainder (584, 34.8%) had no record of donation. Comparing females to males (F/M), there was no significant difference in donor proportion (65.0/65.5%, P = 0.816) or race (62.4/62.0% Caucasian, 9.6/8.3% African American, 3.9/3.9% Other, P = 0.794) (Table 1). Females were older (41.9/39.65yr, P < 0.001), shorter (1.64/1.78 m, P < 0.001) and weighed less (72.4/88.0 kg, P < 0.001), and had slightly lower BMI (26.9/27.6 kg/m2, 2 = 0.003) compared to males.

Table 1 Cohort summary demographics split by sex.
Full size table

Reference values

Sex-specific mean and standard deviation (s.d.) healthy, adult reference values for VFA and VFRA at T10–L4 vertebra levels are reported separately for contrast and non-contrast phase for the reference − 205 to − 50 HU range (Table 2). Peak VFA (contrast/noncontrast) was observed at L4 in females (79.8/84.9 cm2) and L2 in males (141.4/148 cm2) while trough VFA was observed at T11 in females (51.6/53.9 cm2) and T10 in males (98.2/97.1 cm2). Peak VFRA was observed at L1 in both females (− 92.2/− 90.8HU) and males (− 95.5/− 94.3HU) while trough VFRA was observed at T10 in both females (− 97.0/− 98.1HU) and males (− 99.9/− 101.0HU).

Table 2 Male and Female mean and standard deviation for T10-L4 VFA and VFRA split by contrast status, vertebra level.
Full size table

Contrast versus non-contrast

VFA was greater in non-contrast versus contrast scans for all mean paired differences, except for those at T10 in males. In males at T10 the differences were not significantly different for all HU ranges. Across all vertebra levels and all HU ranges, mean paired differences ranged from − 3.0 to − 8.1% in women and 1.6 to − 8.0% in men, expressed as percent of the mean non-contrast VFA (Table 3).

Table 3 Mean VFA paired differences (contrast–non-contrast) raw (\(\mu _d\)) and as percent of overall mean non-contrast value (\(\mu _{nc}\)).
Full size table

The − 190 to − 30 HU range showed the largest paired differences in VFA between contrast and non-contrast scans, followed by − 150 to − 50 HU, − 250 to − 50 HU, − 195 to − 45 HU, and finally − 205 to − 51 HU. The largest differences between contrast and non-contrast were found at L3, L4, and T10 in females and at L3 and L4 in males, across all HU ranges.

VFRA was greater in non-contrast versus contrast scans for all mean paired differences between L1 and L4 in males and females, however, for T10–T12 the results differed by HU range and sex (Table 4).

Table 4 Mean VFRA paired differences (contrast–non-contrast) raw (\(\mu _d\)) and as percent of absolute mean non-contrast value (\(\mu _d/|\mu _{nc}|\)).
Full size table

For both VFA and VFRA, Bland-Altman plots of agreement demonstrated proportional bias at all vertebra levels and in all HU ranges; as the mean value increased there was a trend toward greater differences and the variance of the differences increased (Figs. 1 and 2). For VFA, the trend was positive at T10, neutral/negative at T11, and negative for T12 through L4. For VFRA, trends varied by vertebra level and HU range. Comparing HU ranges, the variance of VFA was fairly consistent across ranges, however, the variance of VFRA increased as the HU range increased.

Figure 1
figure 1

Bland-Altman plot of VFA (cm2) by vertebra number, and HU range showing each observation (black point), mean (blue line) and 1.96 standard deviations (red lines) of the paired differences (contrast-non-contrast), linear best-fit regression (yellow line).

Full size image
Figure 2
figure 2

Bland-Altman plot of VFRA (HU) by vertebra number, and HU range showing each observation (black point), mean (blue line) and 1.96 standard deviations (red lines) of the paired differences (contrast-non-contrast), linear best-fit regression (yellow line).

Full size image

Fat HU range

Compared to the − 205 to − 50 HU reference range, VFA was lower for the − 150 to − 50 HU range, and higher for all other ranges at all vertebra levels in both sexes and for both contrast and non-contrast scans (Tables 5 and 6). The largest differences were observed with the − 190 to − 30 HU range, which ranged from 15.4% (L3) to 20.2% (T10) in females and from 10.8% (L3) to 14.8% (T10) in males for contrast scans (Table 5), and from 17.1% (L3) to 22.2% (T12) in females and from 11.8% (L3) to 17.0% (T10) in males for non-contrast scans (Table 6). The − 195 to − 45 HU range showed the second largest (positive) differences, followed by the − 150 to − 50 HU (negative), and finally the − 250 to − 50 HU range (positive) in both contrast and non-contrast scans.

Table 5 Contrast scan mean VFA paired differences (comparison range\(-\)reference range) raw (\(\mu _d\)) and as percent of overall mean reference range value (\(\mu _d/\mu _r\)).
Full size table
Table 6 Non-contrast scan mean VFA paired differences (comparison range–reference range) raw (\(\mu _d\)) and as percent of overall mean reference range value (\(\mu _r\)).
Full size table

Compared to the − 205 to − 50 HU reference range, VFRA was lower for the − 250 to − 50 HU range, and higher for all other ranges at all vertebra levels in both sexes and for both contrast and non-contrast scans (Tables 7 and 8). The largest differences were observed with the − 190 to − 30 HU range, which ranged from 9.8% (L4) to 12.6% (T10) in females and from 7.6% (L3) to 10.1% (T10) in males for contrast scans (Table 7), and from 10.6% (L4) to 13.6% (T10, T11) in females and from 8.2% (L3) to 11.7% (T10) in males for non-contrast scans (Table 8). The − 150 to − 50 HU range showed the second largest (positive) differences, followed by the − 195 to − 45 HU range (positive), and finally the − 250 to − 50 HU range (negative) in both contrast and non-contrast scans.

Table 7 Contrast scan mean VFRA paired differences (comparison range–reference range) raw (\(\mu _d\)) and as percent of absolute overall mean reference range value (\(\mu _d/|\mu _r|\)).
Full size table
Table 8 Non-contrast scan mean VFRA paired differences (comparison range–reference range) raw (\(\mu _d\)) and as percent of absolute overall mean reference range value (\(\mu _d/|\mu _r|\)).
Full size table

Outlier investigation

Individuals with paired differences greater than 4 s.d. above/below the mean difference were identified and investigated individually as possible outliers. These pairs displayed distinct differences in lung inflation, resulting in vertical shift of organ locations and large differences in axial visceral non-muscular contents between the two phases of scan. Differences were more visually apparent at the thoracic levels where lung air was directly observable, and less apparent in the lumbar region where differences were due to more subtle shifts in major organs including liver, kidney, spleen, and bowel (Fig. 3).

Figure 3
figure 3

Example of healthy 36 y/o male non-contrast (left) and contrast (right) CT axial slices, showing T11–L5 visceral fat area (yellow-shaded region), subcutaneous fat area (purple shaded region), outer abdominal fascia boundary (yellow line), and skin boundary (purple line). Images demonstrate different lung inflation between contrast and non-contrast scans and resulting shift in organ and fat pixels.

Full size image

Discussion

Both the presence of intravenous contrast and the particular HU range used to define visceral fat pixels affected measurements of visceral fat cross-sectional area and mean radiation attenuation on CT.

Analysis of contrast versus non-contrast scans showed that VFA was higher in non-contrast scans compared to contrast scans, whereas VFRA was lower in contrast scans compared to non-contrast scans. The difference in HU was in general quite close to zero, however, Bland-Altman plots demonstrated proportional bias-differences increasing with the mean value-particularly in VFA measurements. The differences due to contrast increased as the area or density of visceral fat increased. Thus, patients with greater visceral fat or greater fat density had greater measurement differences between scans with and without contrast. Therefore, to fairly compare fat measurements for patients with low and high visceral fat, contrast and non-contrast scans should not be mixed within the same study.

Additionally, measurement differences due to HU range were much greater than those due to contrast. In particular, the HU range of − 190 to − 30 showed the greatest differences in both VFA and VFRA. This range was the only range that includes HU values greater than − 45 HU, suggesting that the large difference was primarily due to pixels between − 45 and − 30 HU being counted as fat tissue. Hence, we conclude that visceral fat measurements derived using different fat HU ranges cannot be directly compared.

Outlier investigation showed that reliability of both VFA and VFRA measurements are strongly dependent upon breath cycle; individuals whose paired scans demonstrated large differences in lung inflation also demonstrated large differences in VFA and/or VFRA, particularly in the thoracic region. Caution should be exercised when using visceral fat measurements from the thoracic region as these measurements will be strongly affected by the amount of air held in the lungs at the time of the scan.

This study has important limitations. Our cohort may not be nationally or globally representative, though it is not specific to a particular race or ethnicity. These reference values have not been tested against clinical outcomes and were computed from kidney donor CT protocols only; we did not evaluate the effect of different kVp, mA, convolution kernel, contrast dosage, and/or slice thicknesses used in other protocols nor did we evaluate the effect of various disease states, height, or BMI on these measurements. Because we used retrospective scans, we could not control the slice thickness used in each scan and slice thickness was correlated with the type of scan due to scan protocols used. The vast majority of non-contrast scans used 5mm slice thickness whereas the majority of contrast scans used 2.5mm or smaller slice thickness, resulting in ’sharper’ contrast images and ’blurrier’ non-contrast images. The effect of this difference was not specifically addressed in this manuscript, though visual inspection of distributions showed no large, obvious deviations.

Clinical CT scans obtained in the normal course of patient care can potentially be used for detailed cross-sectional visceral fat evaluation. However, scans may not include vertebral levels for which reference values have been defined (e.g., L3) and may or may not use contrast enhancement. Furthermore, we have shown that the fat HU range used to identify fat pixels can significantly change the resulting measurements, rendering it perhaps one of the most important choices made in CT fat measurement. We report healthy population reference values for a wide range of vertebral levels, in both contrast and non-contrast phases, and using multiple different fat HU ranges. These values provide strong evidence towards standardizing visceral fat area and radiation attenuation measurements and provide a healthy reference population for other studies to compare against.

Methods

Study cohort

We retrospectively studied persons who underwent CT scans at Michigan Medicine as part of evaluation for kidney donation between 1999 and 2017. We have previously studied a subset of these kidney donor candidates as a healthy reference population for skeletal muscle25.

Patient age, sex, height, and weight were obtained from their medical record proximal to the date of evaluation for kidney donation, and the month and year of the evaluation appointment was recorded27. Candidates were included if they had a CT scan performed as part of evaluation for kidney donation, were deemed healthy enough to donate, had age, sex, height, and weight recorded in their electronic medical record, had both contrast-enhanced and non-contrast-enhanced series available, and had a fascia boundary that was fully visible in the display field of view.

Body mass index (BMI) was computed and categorized into groups according the World Health Organization International Classification standard28. Race, unavailable for 25% of the cohort, was reported but not specifically analyzed.

CT imaging was extracted for 2,902 total donor candidates between the ages of 18 and 75. The n = 1,677 candidates of all ages having both a contrast and non-contrast series scan were used in this analysis.

All candidates included in the analysis were scanned using the GE ‘Standard’ reconstruction algorithm (which is optimized for visualizing soft tissue) at 120 kVp in a Discovery or LightSpeed scanner. Non-contrast scans used 2.5 (2007–2010), 3.75 (2008), or 5 mm (1999–2017) slice thickness and contrast scans used 0.625 (2005–2010), 1.25 (2002–2008), 2.5 (1999–2002, 2010–2017), 3.75 (1999), and 5 mm (2002–2016) slice thickness. Tube current was automatically modulated in proportion to body mass.

CT image processing

After being transferred into a spatial database, CT images were processed using Analytic Morphomics, a semi-automated image analysis method that has been previously described24, 29, 30. A combination of automated and user-guided algorithms written in Matlab (The Mathworks Inc, Natick, MA) identified the vertebral bodies to serve as an anatomical coordinate reference system. Next, the outer abdominal fascia and skin boundary were identified at all available vertebral levels to create enclosed regions of interest, which were confirmed by multiple trained researchers (Fig. 4).

Figure 4
figure 4

Example of healthy 20 y/o male non-contrast (left) and contrast (right) CT axial slices, showing T10-L4 visceral fat area (yellow-shaded region), subcutaneous fat area (purple shaded region), outer abdominal fascia boundary (yellow line), and skin boundary (purple line). Portions of skin boundary that are coincident with the scan field of view are highlighted in red.

Full size image

Measurements were taken at each vertebral level between T10 through L4. Sample size at each vertebra level varied due to differences in anatomy included in each subject’s scan. Vertebra levels that did not include both a non-contrast and contrast measurement were excluded.

Fat measurements were computed using the axial slice nearest the inferior aspect of each vertebral body. Visceral fat area (VFA) was computed as the total area of pixels enclosed by the outer abdominal fascia, falling within five different fat Hounsfield Unit (HU) ranges: (1) − 205 to − 51 HU (our reference), (2) − 150 to − 50, (3) − 195 to − 45, (4) − 190 to − 30, (5) − 250 to − 50 HU. Mean radiation attenuation (VFRA) was computed as the mean HU of all pixels included in VFA.

Statistical methods

Male and female demographics, CT parameters, and fat measurements are shown separately as mean +/− s.d. for continuous variables and proportion for categorical variables. Continuous means were compared using two-tailed t-tests assuming unequal variance and categorical proportions were compared using the Chi-squared test.

The within-subject difference between (1) enhanced and non-enhanced fat measures, and (2) between each HU range and the reference range were assessed using paired t-tests. The mean difference and the mean difference expressed as a percentage of the (1) enhanced and (2) reference measurement are reported for each test. Fat HU ranges were analyzed separately in enhanced and non-enhanced scans.

Bland-Altman plots were used to visualize the agreement between pairs of measurements31.

The sex-specific mean and standard deviation of each fat measure were calculated independently for vertebral levels from T10 to L4.

An alpha level of .01 was used to determine statistical significance. All statistical tests were performed in R version 4.0.232, using the package ‘ggplot2’33 for data visualization.

Ethical approval and informed consent

This study was approved by the Institutional Review Board of Michigan Medicine. All methods were performed in accordance with the relevant guidelines and regulations of the United States. Because existing CT scans were used retrospectively, the requirement for informed consent was waived by the Institutional Review Board of the University of Michigan.