Measurement of mean subcutaneous fat thickness: eight standardised ultrasound sites compared to 216 randomly selected sites

Ultrasound (US) provides the most accurate technique for thickness measurements of subcutaneous adipose tissue (SAT) layers. This US method was recently standardised using eight sites to capture SAT patterning and allows distinguishing between fat and embedded fibrous structures. These eight sites chosen for fat patterning studies do not represent the mean SAT thickness measured all over the body that is necessary for determining subcutaneous fat mass. This was obtained by SAT measurements at 216 sites distributed randomly all over the body. Ten participants with BMI below 28.5kgm−2 and SAT means (from eight sites) ranging from 3 mm to 10 mm were selected. The means from eight sites overestimated the means obtained from 216 sites (i.e. 2160 US measurements in the ten participants); the calibration factor of 0.65 corrects this; standard deviation (SD) was 0.05, i.e. 8%. The SD of the calibration factor transforms linearly when estimating the error range of the whole body’s SAT volume (body surface area times the calibrated mean SAT thickness). The SAT masses ranged from 3.2 to 12.4 kg in this group. The standard deviations resulting from solely the calibration factor uncertainty were ±0.3 and ±1.0 kg, respectively. For these examples, the SAT percentages were 4.9(±0.4)% and 13.3(±1.0)%.


Ultrasound Measurement of Subcutaneous Adipose Tissue (SAT)
Adipose tissue is mainly stored in the subcutaneous region of the body, but significant amounts can also be found near organs (visceral adipose tissue, VAT), in the bone marrows, and within tissues, e.g. in muscle 9,22 . The amount of SAT ranges from only a few kilograms up to 50% of body weight and even more 17 . Adipose tissue consists of adipocytes and embedded collagen and elastin fibres which support the tissue 23 .
US was used for SAT thickness measurements in 1965 and 1966 already 24,25 . Bellisari et al. found that interand intra-observer errors were less than 0.15 mm at all of their investigated sites, except for triceps where they found 0.6 mm 26 . US has also been applied for measuring visceral fat. Koda et al. 27 compared MRI measurements to US measurements of subcutaneous and visceral fat and they already found that US was highly accurate and reliable. 2016 a standardised US technique for measuring SAT has been presented 16,17 .
US is the only imaging method capable of measuring both the thickness of the fat layer with and without the embedded structures [14][15][16]28 . The applicability of this novel approach in groups with overweight and obesity has been shown by Störchle et al. 17 . Eight sites are used to represent the trunk (three), the arms (two), and the legs (three). US images of all sites show a simple structure: skin, SAT, muscle fascia. The sites were selected such that the thickness of the layer does not change appreciably in the vicinity of the site; this increases reliability. Site marking is easy and can be learned with high precision within short time (one hour training is sufficient). All distances necessary to define the sites are relative to the body height of the person. This highly accurate and reliable US approach avoids compression artefacts, distinguishes between fat tissue and embedded structures, is not invasive, does not use ionising radiation, and is easily applicable in the field.
However, it cannot be assumed that the mean thickness value of these eight standardised sites that were selected for fat patterning analyses is the best representation of the real mean SAT thickness which is needed to calculate the fat mass. Therefore, extended measurement series containing many more (and randomly selected) sites are necessary to calibrate the mean obtained from the eight standardised sites 16,17 .

Results
In a group of ten male participants (Table 1), subcutaneous adipose tissue (SAT) was measured twice at eight standardised sites using a recently developed ultrasound (US) method 16,17 . As an example, the US image of SAT at one of the eight standardised sites (lateral thigh, LT) is shown in Fig. 1d. In addition, SAT was also measured with the same US technique at 216 sites that were randomly distributed all over the body (Fig. 1a-c). The sums of the eight SAT thicknesses (D) of all 10 participants are shown in Table 2, and mean values of these eight measurements (d M8 ) are presented in Table 3. Thicknesses including the fibrous structures embedded in the SAT are indicated by the index "I", measurements where these structures were excluded are indicated by the index "E", and "F" indicates the thicknesses of the fibrous structures. For SAT thicknesses at individual measurement sites, the lower case letter "d" is used, and for the sums obtained from the eight sites at each participant, capital "D" is used. The measurements at the 216 sites resulted in the reference means of SAT thicknesses for each of the participants (Table 3). A comparison of the SAT means obtained with the eight standardised sites 16,17 is presented in Fig. 2. Means of typically 50 to 300 measurements obtained from each US image were used to represent the SAT thickness at a given individual site.
The mean thicknesses obtained from the eight standardised sites deviated from the means obtained from the 216 randomised sites. The factor k represents this for the individual ten participants (Table 4). For measurements that included (index "I") fibrous structures the calibration equation is: d IM216 = d IM8 • k IM216 , and for measurements that exclude (index "E") fibrous structures: d EM216 = d EM8 • k EM216 . The table also shows the factors when the eight-site measurements are compared to the 108-site measurements (the measurement series of 216 sites was taken in two series of 108 sites each). The k-values corresponding to 216 measurements with fibrous structures included (I), and without (E), are shown in Fig. 3a,c, and for the 108-site measurement series in Fig. 3b,d. Mean k values were 0.65 in both cases. Figure 4a and c show the correlations between the mean thicknesses obtained for the 216 randomised sites (d M216 ) and the calibrated means (d M8,k ) from the eight standardised sites according to: d M8,k = k • d M8 . The correlation coefficient R² was 0.95 (p < 0.01), and SEE was 0.34 mm for data including fibrous structures (Fig. 4a), and R² was 0.94 (p < 0.01), and SEE 0.36 mm without fibrous structures (Fig. 4c). Figure 4b and 16 , and from 216 randomised sites on the body of each of the ten participants. The latter measurements were performed in two series of 108 measurements each (d M108 ). The index I stands for fibrous structures included in the thickness measurement, E for excluded, and F for the thickness of the fibrous structures. Mean SAT thicknesses of the 11 body segments (BS) head, neck, anterior trunk, posterior trunk, upper arms, forearms, hands, buttocks, thighs, legs, an feet (d M,BS ) are presented in Fig. 5a,b and Table 5. Highest mean value was 12 mm at buttocks and lowest was 0.3 mm at hands. The SAT percentages (SAT %,BS ) for each segment are presented in Fig. 5c,d and Table 5. The columns represent the percentages of the SAT volumes (and thus also of the fat mass percentages) of the 11 body segments.

Discussion
A representative mean SAT value is important for correct calculation of the total subcutaneous adipose tissue (SAT) mass. For this purpose, the mean obtained from the eight sites needs to be calibrated by a representative mean obtained from a large number of sites distributed randomly all over the body. The classification of body segments according to Lund and Browder (Fig. 1c) 32 implies to use: 4 sites on the head, 1 on neck, 7 on anterior trunk, 7 on posterior trunk, 3 on buttocks, 4 on upper arms, 3 on forearms, 3 on hands, 10 on thighs, 8 on legs, and 4 on feet. This would amount to 54 (the genital area was neglected). Several series of 54 sites in the same individual (pilot study, not shown) indicated that scattering of the means of the 54 sites was too high for the purpose of this calibration study. When using 108 sites, scattering is still noticeable as can be seen in Fig. 3b,d. Therefore, we used 216 sites corresponding to about one site per dm 2 in adults.
The mean values of the individual calibration factors (Table 4) were 0.65 for both k IM216 and k EM216 (derived from comparisons of 2160 measurements at randomly chosen sites -216 measurements in ten participants -and additional two times eight measurements at the standardised sites; this amounts to 2320 US measurements). Each thickness measurement at an individual site was represented by the mean of typically 100 thickness values (depending on the breath of ROI setting) measured by the evaluation software (amounting to more than 200,000 thickness values). Although there is still a rest of scattering when using 216 sites due to the randomisation, the standard deviation of the mean calibration factor is low because it is obtained from all ten persons which reduces the "randomisation"-scattering. The means obtained from the eight sites for each individual participant in this group of male participants ranged from 2.9 to 10.2 mm, and the corresponding means obtained from the 216 randomised sites ranged from 1.9 to 6.1 mm.
The mean of eight sites used in the standardised ultrasound method for studying SAT patterning 16,17 overestimated the mean obtained from 216 randomly distributed sites in all individual cases (Fig. 2a,b, Table 3). This overestimation is not surprising as the standardised eight sites were developed to investigate the fat patterning of the body and therefore includes some of the main fat depot areas for subcutaneous fat deposition (femero-gluteal region, back, and anterior abdominal wall). These fat depot areas are represented by five of the eight sites (FT, LT, ES, UA, and LA). The mean calibration factor k = 0.65 (derived from values in Table 4; the SD of k was 0.06) is used here to correct this overestimation for both thickness measurements with fibrous structures included Mean SAT thicknesses at eight standardised sites compared to measurements at 216 and at 108 sites. Ten participants were measured at 216 sites (in two series of 108 sites each) distributed randomly all over the body (compare to Fig. 1a-c), and also at the eight standardised sites. The ten participants are ordered according to increasing mean SAT thickness (d IM8 ). The mean SAT thicknesses (d M ) obtained from measurements at 216 (or at 108) sites deviate by a factor k from the means of the eight standardised sites:  Limitations and further developments. 1. Detailed SAT and fat density values of humans as functions of site, temperature, age, and hydration status are missing. Ethics permission for such studies of human SAT in our lab is given already, and these measurements will start in due course. All imaging methods that can be used for fat mass determination based on volumetry will benefit from accurately determined SAT density in humans. 2. Body surface was determined only approximately in this study using DuBois' data and related formulas and not with the high accuracy obtainable with a state-of-the-art 3D scanner because the focus of this study is on minimising the error due to SAT thickness means obtained by US from only eight (standardised) sites.
3. Determination of SAT mass (or pure subcutaneous fat) mass by US as used here does not capture visceral fat. Attempts to assess visceral fat by US have been made by other groups 37,34 , however, US only detects surrogate parameters like intra-abdominal distances. MRI is capable of measuring visceral fat, although fat layer thickness measurements do not reach the accuracy of US thickness measurements because the pixel size is typically 1.3 to 2 mm in total body scans, and MRI measurement sequence and image segmentation protocols are not standardised for this purpose. Development of improved MRI methods towards higher standards for fat studies is in progress in our laboratory. Comparisons of SAT (measured by US) with VAT (measured by MRI) will show the possibilities and limitations of total body fat (TBF) assessment (on the anatomical level) based on US SAT measurements solely. We assume that there is a good chance to find useful correlations for acceptable assessment accuracy because SAT, which can be determined with high accuracy by US, accounts for typically about 80% of TBF [35][36][37] and therefore scattering of the VAT percentage can be expected to have minor effect on the TBF assessment error. However, there may be outliers, particularly in groups with obesity or extreme underweight. Meanwhile, we use the waist to height ratio (W = w/h), which is the most important anthropometric predictor of health and premature death caused by obesity 38 , as a surrogate measure for VAT. The studies in progress will also show whether a combined approach (US SAT measurements and anthropometric indices like W) will improve the assessment accuracy. 4. In our group of ten male participants, BMI ranged from 20.0 to 28.4 (Table 1), and total SAT percentage from TSAT I% = 4.9% to 13.3% (mean was 8.6% of body mass; Table 2), which represents a group ranging from normal weight to overweight (according to the WHO definitions) 39 . In groups with underweight, SAT thickness of fat depots may get closer to the SAT thickness of other sites; preliminary studies indicate that a higher calibration factor can be expected in such cases. The investigation presented here includes young white Caucasian males. Therefore, similar studies with females, and with other ethnic groups will be of interest, although it would be surprising when results deviated substantially in such cases because the eight sites cover a representative set of fat depot sites, and differences from site to site in different groups can be expected to equal out to a large extend when means of all eight sites are taken. A future focus should also be on older adults and on children, who are not included here.
5. Eight representative sites were chosen because the development of the US method started out from the eight sites that have been used since many years by ISAK (International Society for the Advancement of Kinanthropometry) for skinfold measurements 14,15,28 . It turned out that several of these ISAK sites are not well suited for US measurements (because of complex underlying anatomical structures that are difficult to interpret), and that marking of these sites is difficult and time consuming. Therefore, most of them had to be replaced by new ones; however, the number of eight sites remained. Data mining studies will show whether this number or possibly a reduced number of sites will be the optimum choice for SAT (and for TBF) mass assessment based on US.  Table 3

Methods
Participants. 50 young male participants with BMI values below 28.5 kgm −2 were investigated at the eight standardised US sites and ten were selected out of this set to cover a range of mean SAT thicknesses (from eight sites) from approximately 3 mm to 10 mm (with fibrous structures included: d IM8 ). According to a preliminary schedule and discussion basis 40 for defining SAT ranges, this group represents the SAT categories "desirable range" (2.5 to 7.5 mm) and "noticeable ballast weight" (d IM8 from 7.5 to 12.5 mm) for male persons of the general public. For competitive male athletes, according to the above-mentioned preliminary schedule, the desirable range is 2.5 to 3.8 mm, noticeable ballast weight 3.8 to 6.3 mm, and considerable ballast weight is above 6.3 mm). For anthropometric data see Table 1. Permission for the study was provided by the ethics committee of the Medical University of Graz (20-295 ex08/09).

Informed consent. All participants received an information letter and completed a written consent form.
Standardised sites for US measurements. The standardised eight sites described by Müller et al. 16  Series of 216 SAT measurements in each of the ten participants (two series of 108 measurements each). The 216 sites were randomly distributed all over the body (Fig. 1a-c). Eleven body segments ( Fig. 1c) were covered with a number of sites proportional to their contribution to the total surface area. The body segments were chosen in accordance with the criteria of Lund and Browder 32 , but without the segment "genitalia" (which was ignored here). The surface areas of the head, hand, and foot correspond to 7%, 5%, and 7% of the total surface area and should therefore be represented by 16,12, and 16 measurements, respectively. However, it is extremely difficult to find such a high number of useful sites on hands and feet because of many vessels and complex anatomical structures there. Therefore, only half of the corresponding site numbers were measured and these values were considered twice. As it is very inconvenient for the participant to measure 16 sites on the head, therefore the same approach was used for the head too.    Fig. 5(a,b). The mean percentages (of all ten participants) of the fat mass contributions of the individual body segments are also shown (SAT%,BS); this sums up to 99% (genitalia, wich represent 1%, were ignored). Ultrasound imaging of SAT. US imaging is based on the pulse-echo technique. A series of US pulses (each several wavelengths long) is sent into a given tissue. Medical diagnostic US systems conventionally use c = 1540 ms −1 for distance (d US ) determination in soft tissue. In adipose tissue, the speed of sound is lower: 1450 ms − 1 16,33,41 , and therefore, this appropriate speed of sound is used for thickness measurements in this study (where mean SAT thicknesses ranged from 1.9 to 6.1 mm). Goss et al. 42 list speed of sound measurements (performed in 1953) in human fat: for SAT in which connective tissue was removed, 1459 ms −1 were measured, i.e. a value that does not deviate noticeably from the value of 1450 ms −1 used here 33 : For a 6 mm thick fat layer, a difference of 10 ms −1 would result in a distance measurement error of only 0.04 mm, which is far below the measurement error due to the limited US resolution. In thick layers, the correct choice of sound speed can noticeably influence the measurement error 17 , however, for example in a fat layer as thick as 30 mm, the error due to a sound speed difference of 10 ms −1 is only 0.2 mm, which is still within the range of the error determined by the image resolution. In this US method, the thicknesses d E measure SAT without embedded fibrous structures, whereas d I -values include these connective tissues. Values of the speed of sound in connective tissue embedded in SAT are not available, but even a speed of sound of 1500 ms −1 (for comparison, in the arterial wall tissue 1501 ms −1 were measured 42 ) would not noticeably effect measurement accuracy: assuming 50 ms −1 sound speed deviation (from the used 1450 ms −1 ), about 0.2 mm distance difference would result in a 6 mm thick SAT layer (in human SAT, typically about 15% is connective tissue, and this percentage decreases with increasing SAT thickness). Such possible measurement errors cannot be corrected currently because the speed of sound in the particular connective tissue embedded in SAT is not known yet. Again, these possible errors are small compared to the reliability errors and when compared to the furrowed borders of SAT. All US measurements, at both the standardised eight sites and at the randomised 216 sites, were performed according to the standardised approach 16,17 . All US images clearly showed the lower margin of the skin and the upper margin of the muscle fascia (Fig. 1d). To avoid compression of fat, the probe was placed above a given site by using a thick layer of US gel between the probe and the skin (typically about 5 mm). Measurements were carried out using a GE Logiq-e US device and the linear probe L8-18i operated at 9 to 18 MHz.
Interactive multiple thickness measurement of SAT layers. A semi-automatic evaluation algorithm designed for multiple interactive evaluations of SAT layer thicknesses was used to evaluate the mean SAT layer thickness in a given US image (USTissue Scientific -FAT Analysis Tool, www.iasms.org, www.rotosport.at). The number of thickness measurements in each individual image was typically 50 to 300 (depending on the selected region of interest breadth). In the evaluation software, sound speed was set to 1450 ms −1 for distance calculation in adipose tissue 16,33 . According to the standardised protocol 16,17 , the tissue segmentation was controlled visually and, if necessary, parameters that determine the accepted segment inhomogeneity were set manually to optimise the SAT contour detection. The software enables the operator to distinguish between distances in which fibrous structures are included (index "I") or excluded (index "E"). Figure 1d shows an example of an evaluated US image. The centre lines in the image correspond to the centre of the US probe which was held exactly above the marked site. The rectangular ROI was usually set symmetrically to the centre line. The contour detection algorithm starts out from the manually set circles (or ellipses) and measures multiple thicknesses automatically. Using a ROI that is set symmetrically to the centre line of the US image ensures that the thickness is determined at the marked site (the middle of the US probe is positioned at the marking on the skin, with a thick layer of gel in-between which avoids compression artefacts). A visual control makes sure that the algorithm detected the SAT layer correctly. Accuracy of US thickness measurements. Diffraction and technically obtainable minimum pulse length limit lateral and axial resolution approximately to the wavelength used. Diagnostic US probes (transducers) use frequencies from 3 to 22 MHz, which corresponds to a wavelength in soft tissue of 0.5-0.07 mm. US attenuation increases with increasing frequency-typical investigable depths are between 10 mm (22 MHz) and 200 mm (3 MHz). In this study, 18 MHz were used for thin layers with a border detection resolution of about 0.1 mm, and for thick fat layers, lower frequencies down to 9 MHz were used (resulting in a border detection resolution of about 0.2 mm). There is no relevant dependency of sound velocity on frequency for diagnostic US measurements of tissue thickness 43 . The calculated distance is: sound speed (in SAT) multiplied by half the echo time. The technical measurement error can be kept very low (about 0.2 mm at 18 MHz). The limiting accuracy factors are of biological nature because they are beyond the technically obtainable image resolution and measurement accuracy: furrowed tissue borders and viscous-elastic behaviours of SAT. The influence of these biological limitations is minimised because the image evaluation algorithm used takes mean values of many thickness measurements (typically about 100) in a given image which results in a standard error of the mean that is a magnitude lower than the standard deviation of the individual measurements 17 . A sound speed deviation from the "real" value of 10 ms −1 would result in 0.07 mm thickness difference when measuring a SAT layer of 10 mm, for example. Temperature of fat tissue underneath the skin (SAT) may range between about 30° and 35°. As the human body consist of about 70% of water, the speed of sound in soft tissue is largely determined by the speed of sound in water. For water, the temperature coefficient is well known 44 : it is about 0.14% per K. In other words: A temperature difference of 5° (and more is not to be expected when measurements of humans are made in room temperature, typical 23° to 25° in our laboratory) would result in 0.7%, i.e. a change from 30° to 35° would change the speed of sound from 1509 to 1520 m/s 44 . There are no data available that would give a reason to anticipate substantially larger effects in SAT. Detailed discussion of accuracy limitations have been published previously already 15-17 . Reliability of US measurements. To obtain maximum reliability, it is paramount to mark sites and capture images in accordance with the standardised US measurement technique 16,17 . In a recent publication, an inter-observer study was performed (three observers measured 12 athletes) where mean values of the eight Statistics. For statistical analysis, SPSS (Version 24) was used. Normal distribution was tested by the Shapiro-Wilk test. Statistical analyses included the determination of standard errors of the estimate (SEE), linear regressions including coefficients of determination (R²), and significances (p < 0.01), and limit of agreement (LOA) 45 .

Data Availability
All relevant data generated or analysed during this study are included in this article.