Multiple measures derived from 3D photonic body scans improve predictions of fat and muscle mass in young Swiss men.

INTRODUCTION: Digital tools like 3D laser-based photonic scanners, which can assess external anthropometric measurements for population based studies, and predict body composition, are gaining in importance. Here we focus on a) systematic deviation between manually determined and scanned standard measurements, b) differences regarding the strength of association between these standard measurements and body composition, and c) improving these predictions of body composition by considering additional scan measurements.
METHODS: We analysed 104 men aged 19-23. Bioelectrical Impedance Analysis was used to estimate whole body fat mass, visceral fat mass and skeletal muscle mass (SMM). For the 3D body scans, an Anthroscan VITUSbodyscan was used to automatically obtain 90 body shape measurements. Manual anthropometric measurements (height, weight, waist circumference) were also taken.
RESULTS: Scanned and manually measured height, waist circumference, waist-to-height-ratio, and BMI were strongly correlated (Spearman Rho>0.96), however we also found systematic differences. When these variables were used to predict body fat or muscle mass, explained variation and prediction standard errors were similar between scanned and manual measurements. The univariable predictions performed well for both visceral fat (r2 up to 0.92) and absolute fat mass (AFM, r2 up to 0.87) but not for SMM (r2 up to 0.54). Of the 90 body scanner measures used in the multivariable prediction models, belly circumference and middle hip circumference were the most important predictors of body fat content. Stepwise forward model selection using the AIC criterion showed that the best predictive power (r2 up to 0.99) was achieved with models including 49 scanner measurements.
CONCLUSION: The use of a 3D full body scanner produced results that strongly correlate to manually measured anthropometric measures. Predictions were improved substantially by including multiple measurements, which can only be obtained with a 3D body scanner, in the models.

1. Introduction

Over the last four decades obesity has nearly tripled worldwide and has reached the level of a global pandemic [1-4]. High fat mass, especially in the abdomen (visceral fat), and in connection with obesity, is associated with several diseases, such as coronary heart disease, diabetes mellitus type II and some types of cancer, as well as with all-cause mortality [5-12]. High muscle mass, on the other hand, appears to be beneficial for health and is associated with a reduced risk of functional impairment, disability and mortality, particularly later in life [13-15]. Since fat and muscle mass have such contrasting health implications, measurements of body composition are increasingly important in clinical practice as well as in medical research.

However, there is currently no universally suitable method to measure body composition. Each technique has advantages and disadvantages and its use is therefore highly situational. Standard imaging methods (DXA, CT and MRI) allow body composition to be assessed with high precision by distinguishing between fat and fat-free mass [16-19]. However, these techniques are time-consuming, expensive and/or invasive, and therefore inadequate for study settings in which many probands have to be examined with minimal health risks and in a short period of time [16,17,20]. Bioelectrical Impedance Analysis (BIA) is often used in such settings [16,21,22] because of the easy handling, high measuring speed and transportability of the measuring device. Nevertheless, BIA has some limitations. It is less precise than standard imaging methods [16,21,22], and provides numbers but no visualisation of fat tissue or any record of visible body characteristics. Thus, the results of BIA may seem abstract to lay people.

Despite their inaccuracy, classical manual anthropometric measurements are still most widely used both to estimate a person’s fat mass and the associated health risk in epidemiological studies, and in clinical practice [23]. The most common anthropometric measurements associated with fat mass are Body Mass Index (BMI), Waist Circumference (WC) and Waist to Height Ratio (WHtR) [21]. All these measurements have been shown to be associated with body composition and are therefore able to predict fat mass [22,24,25]. These measurements only require minimal equipment and are based on visible body characteristics with intuitive meaning. They do, however, vary in their significance and precision in different population groups [26], and each of the measures has limitations. The BMI, for example, does not distinguish between fat mass and muscle mass/fat free mass and does not take account of fat distribution [22]. Measurements based on waist girth (WC or WHtR) have therefore gained popularity and seem to be better predictors for fat mass (particularly visceral fat) than the BMI [21,22,24,25,27]. However, these measurements and proportions provide no information on the composition of extremities, where the relative amounts of fat and muscle mass may vary greatly depending on a person’s physical activity, or due to muscle loss in elderly people. Thus, changes in body composition cannot be assessed. Also, interobserver variability in anthropometric measurements may be an issue [23].

Variation in whole-body composition is probably too multidimensional to be properly measured by single distance measurements. As a new procedure to capture this multidimensionality, 3D photonic full body scans can be used [28-36]. Measurements are non-invasive, safe, rapid and simple, and devices are transportable, although less easily than BIA devices. The 3D body scanner functions via photonic scanners surrounding the body. Within 12–15 seconds, millions of data points are gathered, creating a precise 3-dimensional body shape map with detailed body surface topography, from which about 150 standard measures can be derived automatically. These include circumferences of specific body parts, linear dimensions, cross-sectional areas, surfaces, segmental volumes, and proportions, all of which can be used, individually or in combination, to predict health-relevant parameters such as body composition [30,36-40]. Even the established simple predictors of body fat mentioned above (WC, WHR and WHtR) may be measured more rapidly and reliably using the body scanner than manual methods because the improved standardization reduces inter-observer variability [28,30,33,34,37-42].

However, some methodological questions still need to be addressed before 3D body scans can be implemented as a standard method for assessment of body composition. First, it is necessary to check whether manual and scanner-derived anthropometric measures are exactly comparable or whether they differ systematically, so that standard definitions of health risk classes (e.g. increased health risk with a WC greater than 94 cm [26]) must be adapted for scanner-derived measures. As regards the potential use of multiple measures for a more precise prediction of body composition, previous studies only considered a limited number of pre-defined measures [42], and only few of them used automatic variable selection procedures to identify the best predictors [43-45]. Because some of the 150 standard measurements are strongly correlated among each other, model selection procedures and other techniques such as 3D surface geometry may have to account for these correlations [46-51]. Still, further research is needed to identify which of the 150 standard measurements are most relevant for the prediction of body composition, or whether multiple (and partly strongly correlated) measurements are relevant, and how they should be selected or combined to obtain the most reliable predictions.

In this study we analysed a cross-sectional sample of 104 young men and asked the following research questions: Are standard anthropometric measurements assessed manually and by the scanner differently associated with body composition (fat and muscle mass) as estimated by BIA)? Are these predictions of body composition (as estimated by BIA) improved by considering additional measurements provided by the scanner? How many measurements should be included in a multivariable model to obtain the most precise predictions of body composition?

2. Material and methods

This study was part of a larger project in which we examined 104 recruits at the beginning of their Armed Forces basic training [52]. The cross-sectional baseline examinations involved young males (age range 18.8–24.4 years, mean 20.5 years, SD = 1.1 years) recruited by the Swiss Armed Forces and were conducted in Kloten (Canton of Zurich) from 21 March to 24 March 2017. Participation was voluntary and regarding socioeconomic status, origin or other demographic factors, and participation was voluntary. All young men were Swiss nationals (a precondition to be conscripted for mandatory Service for the Armed Forces), but information about migration background or ethnicity was not systematically collected in the questionnaire. Before beginning the study the participants were informed twice about its content and procedure, first in writing and then orally. In addition, informed consent was confirmed in the form of a signature. The ethics committee of the Canton of Zürich formally approved this study (No. 2016–01625).

Because in the setting of the presented study (limited time available for measurements within normal army operations) it was not feasible to perform invasive and more time-consuming examinations, bioimpedance analysis was used to estimate whole body fat mass, visceral fat mass and skeletal muscle mass. The device used was a medical 8-point body composition analyzer (Seca mBCA 515, Seca AG, Reinach, Switzerland), which was validated in several studies and has often been used to compare body composition measures obtained through different measurement methods including 3D body scanners [53-61]. The participants stood barefoot on the four foot-electrodes and grasped the four hand-electrodes with their hands. Alongside the analysis of the body composition, selected anthropometric measurements which are relevant in a medical and epidemiological context were taken manually according to WHO guidelines [26]. Waist circumference (WC) was measured with a hand held-tape measure with stretch resistant quality and automatic retraction (Seca 201, Seca AG, Reinach, Switzerland). Participants were measured at the midpoint between the lowest point of the ribcage and the highest point of the pelvis bone, always by the same trained and experienced researcher. Height and weight were measured with a standard stadiometer (Seca 274, Seca AG, Reinach, Switzerland). The participants wore underwear and stood straight with their feet together.

For the 3D body scan a semi-mobile Anthroscan VITUSbodyscan body scanner was used. Four lasers and eight cameras create a point cloud via optical triangulation containing 300 data points per cm3. The software (Anthroscan 2016, Version 3.5.3) then calculates 150 standard measurements (ISO 7250 / ISO 8559 and DIN EN ISO 20685) including various girths and body part volumes. These body volume estimations (also for body regions) have been shown to be important for relative body fat mass [62] and of good validity and reliability in other studies [63,64]. For this study we included 90 standard measurements as delivered by the software (a complete list with measurement ID numbers is provided in S1 Table). The non-selected measurements were excluded before the start of data analysis based on two criteria: a) Specific measures intended for the textile sector (e.g. for shirts). b) Clearly redundant measurements (e.g. several nearly identical measures for leg length). The scanner was calibrated daily before use according to the manufacturer’s instructions. The participants were scanned wearing tight-fitting underwear in standard position defined by the manufacturer of the 3D body scanner (standing up straight, feet positioned ca. 30 cm apart, arms slightly bent at the elbow and held slightly away from the body, head in accordance with the Frankfurt Horizontal Plane) and held breath after exhalation. To ensure the right positioning, we briefed every participant in advance. Participants wore form-fitting underpants and a tight-fitting bathing cap. Regarding postprocessing of the scans: We worked with the raw point clouds for the extraction of all standard measurementsexcept for the volumes. All 104 scans were checked for their quality (absence of artifacts). For the calculation of the partial volumes, we automatically calculated closed surfaces using the standard procedure in the Anthroscan software (good quality level, medium mesh size), the cutting off of the partial volumes was performed fully automatically via the Anthroscan software, but supervised for quality.

Statistical methods

The agreement between manual and scanned anthropometric standard measurements was assessed through Bland-Altman plots, i.e. by plotting the difference between the two measurements against their mean value for each participant [65,66]. Smoothing lines in these plots showed whether one method yielded systematically higher values than the other, and whether this discrepancy affected the entire range of measured values or only part of the range.

The association between scanned anthropometric standard measurements (BMI, WC, WHtR) and body composition (absolute and relative fat mass, visceral fat mass, and skeletal muscle mass (AFM, RFM, Visc, SMM)) was assessed by Spearman rank correlations and scatter plots with smoothing lines. These plots showed approximately linear relationships between absolute or relative fat mass and each of the predictors, and clearly segmented relationships for visceral fat mass, which was only linearly related to the measurements above a certain threshold. Accordingly, either linear regression or segmented regression was used to compare body fat predictions obtained with manual and scanned standard measurements by computing both the fraction of variation explained (r2), and the prediction standard error, i.e. the square root of the mean squared prediction error obtained by leave-one-out cross-validation. We chose the cross-validation method over method of the splitting the data set in to training and validation data sets because of rather small overall sample size.

The possible gain in predictive value obtained by considering scanned anthropometric measurements other than the standard ones (BMI, WC, WHtR) was assessed by stepwise forward model selection using the AIC criterion. Of the three standard anthropometric measurements, only WC was considered here because it proved to be the best predictor for the three measures of body fat content. The first step of model selection showed whether any of the other 89 scanner measurements would predict body fat content better than WC. Further steps showed how much the prediction could be improved by adding a second, third or more predictors. For easy interpretation, the gain in predictive value was described as the fraction of variation in additional body fat content that was explained when a predictor entered the model. Because some of the 90 scanner measurements were strongly correlated with each other, we expected model selection to be partly arbitrary and determined by random structures in the data. To assess the resulting uncertainty in the choice of the best predictors, we repeated model selection for 2000 bootstrap samples of the data and recorded the first six predictors selected with each sample. We then determined how often individual measurements were selected in the first step, and how often each of the measurements initially selected among the six top predictors were also among these in the bootstrap samples.

Because stepwise model selection tends to overfit the data and produce unreliable solutions when predictors are strongly correlated, we also performed model selection with the lasso procedure. This involves fitting a multiple regression model with a penalized least squares criterion so that most of the unimportant and/or correlated predictors have a coefficient of zero and are dropped from the model. The optimal penalty term was selected by cross-validation using the “minimum + 1se” rule. We compared the predictions obtained with both model types in terms of explained variation (r2), and prediction standard error from leave-one-out cross-validation.

Finally, we fitted a single multivariate lasso model to the four measures of body composition to obtain a single set of scanner measurements that would jointly provide the best predictions for the four body composition measures. We standardized both the scanner measurements and the four composition measures to a mean of 0 and standard deviation of 1 so that we could directly compare the regression coefficients and thus, the relative contribution of each of the selected scanner measurements to the prediction of each measure of body composition.

All analyses were performed using R version 3.5.2 (2018, The R Foundation for Statistical Computing, Vienna). To obtain the Bland-Altman plots we used blandr, the segmented regression was determined using the segmented package, and Lasso models we obtained using glmnet.

3. Results

The descriptive statistics for all manual and scanner measurements are reported in S1 Table. According to standard definitions of BMI categories, 20.2% of the participants were overweight (BMI 25.0–29.9kg/m2) and 5.8% obese (BMI> = 30.0kg/m2). According to the WC, only 4.8% of the participants showed increased disease risk (WC 94-102cm) and 3.8% very high disease risk (WC>102cm), whereas the WHtR suggested that 17.3% had an increased disease risk (WHtR 0.5–0.6) and 1.0% a very high disease risk (WHtR>0.6). The three scanned anthropometric measurements (BMI, WC, WHtR) strongly and positively correlated to each other (Spearman Rho >0.89) (S1 Fig). Visceral fat mass, AFM and RFM were also strongly and positively correlated to each other (Spearman Rho >0.79), whereas the correlations with SMM were weaker (Rho 0.31–0.58) (S2 Fig).

In terms of agreement between methods, scanned and manually measured height, WC, WHtR, and BMI were strongly correlated (Spearman Rho>0.96) (Fig 1). However, the Bland-Altman plots for height showed a constant bias of -1cm towards scanned height being shorter, which resulted in slightly higher BMI values from the scanner. For WC and WHtR there was a trend towards values in the upper part of the range in the scanner than when manually measured.

Fig 1

Agreement between methods: Scan vs. manual by scatterplots (left) and Bland Altmann plots (right) for height (A,B), WC (C,D), WHtR (E,F) and BMI (G,H).

Generally, scanned and manually measured values are strongly correlated (Spearman Rho>0.96). For height there is a constant bias of -1cm towards scanned height being shorter. For WC and WHtR there is a trend towards higher values being larger in the in the scanner than when manually measured.

The associations between the scanned anthropometric measurements for excess weight (BMI, WC, and WHtR) and visceral fat mass, AFM, RFM, and SMM are reported in Fig 2 and Table 1. In general, explained variation and prediction standard errors were similar between scanned and manual standard measurements. The highest explained variation (r2) was observed for AFM and the lowest for SMM. Visceral fat mass showed segmented associations with all anthropometric standard measurements (breakpoint for WC = 78.4 cm). Overall, WC explained more variation than the two other anthropometric standard measurements.

Fig 2

Relationships between body composition (fat or muscle mass as determined through bioimpedance analysis) and three standard anthropometric measurements determined either with the 3D body scanner or through manual measurements.

Segmented regression was used for visceral fat, and linear regression for the other body composition measures. The fit of each regression model is given as explained variation (r2), and its predictive value is given as the prediction standard error, i.e. the square root of the mean squared prediction error obtained by leave-one-out cross-validation.

Table 1

Comparison of univariable and multivariable regression models for the prediction of body composition (fat or muscle mass as determined through BIA) from anthropometric measurements.

In the univariable models, the four measures of body composition were related to three standard anthropometric measurements (BMI = body mass index, WC = waist circumference, WHtR = waist-to-height ratio, determined either with the 3D body scanner or through manual measurements) using linear regression or (for visceral fat) segmented regression. In the multivariable models, the four measures of body composition were related to 87 scanned measurements, from which the relevant predictors where selected either through stepwise forward model selection (to minimize the AIC) or through the lasso procedure. The fit of each model is given as explained variation (r2), and its predictive value is given as the prediction standard error, i.e. the square root of the mean squared prediction error obtained by leave-one-out cross-validation.

	Univariable models (standard measurements)						Multivariable models
	BMI		WC		WHtR		Stepwise selected	Lasso
Visceral fat (kg)	scanner	Manual	scanner	manual	scanner	manual	p = 49	p = 6
Explained variation (r2)	0.76	0.76	0.87	0.92	0.83	0.88	0.987	0.834
Prediction standard error (kg)	0.48	0.47	0.36	0.27	0.43	0.35	0.250	0.429
Absolute fat mass (kg)							p = 19	p = 15
Explained variation (r2)	0.84	0.85	0.87	0.86	0.78	0.77	0.978	0.947
Prediction standard error (kg)	3.38	3.36	3.01	3.19	4.03	4.13	1.638	2.435
Relative fat mass (%)							p = 39	p = 10
Explained variation (r2)	0.77	0.77	0.79	0.78	0.74	0.74	0.975	0.888
Prediction standard error (%)	3.71	3.65	3.54	3.63	3.89	3.90	2.183	2.798
Skeletal muscle mass (kg)							p = 23	p = 19
Explained variation (r2)	0.54	0.53	0.49	0.50	0.28	0.28	0.971	0.943
Prediction standard error (kg)	2.63	2.65	2.77	2.74	3.28	3.30	0.909	1.180

p = number of predictors selected.

Among the scan parameters, circumferential measurements in the abdominal and hip area were highly correlated with relative fat mass (Spearman Rank correlation Rho >0.8). Partial volumes had the highest correlations with skeletal muscle mass (Rho >0.8) (S1 Table). Vertical length and distance measurements showed generally showed weaker correlations with relative fat mass and skeletal muscle mass (Rho <0.4). As expected, predictors belonging to the same measurement type were positively correlated with each other:The average Spearman rank correlations (Rho) of predictors within groups were 0.71 forvertical distances, 0.73 for girths, and 0.79 for partial volumes. Associations among individual scan features are further illustrated by a tree from cluster analysis in S1 Fig.

Stepwise forward model selection confirmed that either WC or a closely related measurement (e.g. belly circumference or maximum belly circumference, high hip girth) was the single best predictor of body fat content (Table 2). In the bootstrap samples, WC was selected most often as a predictor of visceral fat mass, while belly circumference was selected most often as a predictor of AFM and RFM. The inclusion of a second predictor into the model increased the explained variation by 1.2% to 3.2%, and a third predictor explained a further 1.1% to 1.6%. Another 2.0% to 2.5% of variation was jointly explained by predictors 4 to 6. However, most of these predictors were selected among the top six predictors with fewer than 50% of the bootstrap samples, meaning that other measurements could be selected as well. Forearm volume (left or right) was most often selected as the best predictor of SMM, with various measures of leg size an alternative or second predictor, indicating that SMM was mainly related to total limb volume.

Table 2

Detailed results of stepwise forward model selection for the prediction of body composition (fat or muscle mass as estimated through bioimpedance analysis) from scanned anthropometric measurements.

For each of the six measurements selected first in the stepwise procedure, the (additional) fraction of variation in body fat or muscle mass explained by the inclusion of this predictor in the model is given. The stability of model selection was evaluated by running the procedure on 2000 bootstrap samples. For each of the six measurements initially selected first, the fraction of bootstrap samples where this measurement was also among the first six predictors selected is given. In addition, all measurements that were selected in the first step at least once are given.

Variables	% expl.	Among first six (%)	Alternatives for the first (main) predictor (% of bootstrap samples where the variable was selected in the first step)
Visceral fat (kg)
WC	81.8	60.0	WC (52.3), Belly circumference (18.3), High hip girth (13.1), Middle Hip (10.3),
Volume Forearm Right	3.2	26.7	Maximum belly circumference (5.2), High waist girth (0.85), Waist band (0.05)
Middle Hip	1.6	32.0
Distance waistband knee	0.8	17.8
Upper arm girth right	0.7	20.4
Upper torso torsion	0.5	11.3
Absolute fat mass (kg)
Maximum belly circumference	90.9	31.6	Belly circumference (46.5), Maximum belly circumference (31.3), High hip girth (17.0),
Distance waist knee	1.2	12.7	Middle hip (4.6), WC (0.25), X_overview Volume (0.25), Buttock girth (0.05), Hip girth (0.05), Thigh girth right horizontal (0.05), Waist band (0.05)
X_Overview Volume	1.1	56.0
Knee girth left	1.1	32.9
Volume Forearm Left	1	45.6
Forearm girth right	0.5	8.8
Relative fat mass (%)
Belly circumference	83.3	88.5	Belly circumference (88.5), Maximum belly circumference (9.5), High hip girth (1.75),
Thigh girth right horizontal	3.1	35.5	Thigh girth right horizontal (0.25), WC (0.2), Buttock girth (0.15), Hip girth (0.15), Thigh girth left horizontal (0.10)
Volume Forearm Left	2.3	60.1
Dev. waist band from waist back	1	11.3
min leg girth left	0.8	24.3
Elbow girth right	0.5	11.6
Skeletal muscle mass (kg)
Volume Forearm Right	78.5	63.2	Volume forearm right (49.5), Volume forearm left (20.5), Volume lower Leg Right (8.6), X_Overview Volume (8.5), Volume Thigh Left (5.0), Forearm girth left (4.0),
Volume Thigh Left	6.9	23.1	Hip thigh girth (1.4), Volume Lower Leg Left (1.1), Total torso girth (0.85), Buttock
Waist to buttock height left	2.7	9.9	girth (0.15), Elbow girth right (0.15), min. leg girth left (0.15), calf girth right (0.10), Elbow girth left (0.05), Forearm girth right (0.05), min. leg girth right (0.05)
Neck height	2.2	2.4
Forearm girth left	1.8	42.6
Upper arm diameter left	0.9	21.0

The total number of predictors selected by stepwise forward model selection ranged from 19 (AFM) to 49 (visceral fat mass). Both model fit (r-squared) and predictive value (cross-validated r-squared) increased or remained stable up to this large number of predictors (Fig 3). Model fit reached values close to 100%, and predictive value reached more than 90% for the four body composition measures. For the three measures, the mean prediction error of the stepwise selected model, as determined by cross-validation, was small and only slightly larger than the model’s residual standard error despite the large number of predictors included (Table 1). The lasso procedure selected models with less predictors, ranging from 6 (Visc) to 19 (SMM), and with slightly lower predictive value (Table 1). Overall, we only found a moderate degree of overfitting (i.e., only small differences between r2 and cross-validated r2) even with multiple predictors.

Fig 3

Increase in model fit (r-squared) and predictive value (cross-validated r-squared) with increasing number of predictors in the model.

Predictors were included in a stepwise forward selection procedure to minimize the AIC. Curves illustrate how many predictors were needed to obtain the maximal predictive value and the moderate degree of overfitting (difference between r2 and cross-validated r2) found even with multiple predictors.

4. Discussion

In this study we compared estimated body composition (fat and muscle mass) with external body measurements obtained either manually or with a 3D body scanner in a cross-sectional sample of young Swiss men. We found that standard body measurements obtained with both methods were strongly correlated, yet some systematic differences existed. In general, standard measurements obtained with both methods performed equally well in predicting variation in estimated body composition. Of the 90 measurements obtained from the 3D body scans, the single best predictor of body fat was waist or belly circumference, while skeletal muscle mass was best predicted by limb size (length, girth or volume). The inclusion of additional measurements into multiple regression models increased the predictive value of each of the four body composition measures by more than 90%. Stepwise forward variable selection returned models with substantially more predictors than the lasso procedure, yet no overfitting was apparent, and a similar predictive value was achieved with both model selection procedures. However, due to strong correlations between some of the measurements, the exact choice of predictors in the best predictive model was largely arbitrary. Moreover, the optimal prediction function (possibly considering further derived features) still has to be determined in future studies.

The finding of a systematic bias between scanned and manually assessed data is supported by several other validation studies with similar results [42]. In the scanner participants stand with their legs hip-width apart (to enable the scanner/software to correctly identify the crotch), while they stand with their legs closer together when being manually measured [26]. In an earlier study with a different study population [40] as well as in the 54 follow-up assessments of the present study population [46] we found that height was systematically shorter in the scans. The positioning of the legs (hip-wide apart in the standard scan position vs. legs together and straight posture as being manually measured by an anthropometer) was found to be only partially responsible for the systematic height difference. The remaining difference could be related to the fact that we ran the automatic measurements with the software on the raw scans and point cloud might be slightly fragmented on the top of the head and the bottom of the feet. In other studies, height was systematically greater in the scans and that was mainly explained by issues related to the worn bathing cap (air beneath, or lots of hair up-biasing height) [34,42]. Future studies should therefore examine the partly conflicting results regarding the systematic height bias more closely, especially since the calculation of the BMI depends on it. Like in the 54 follow-up assessments of the present study population [46], WC was larger in the scans, which is likely due to the tendency for hand-held tape measurements to compress the waist circumference, caused by too much tension, and thus reduce the values [64]. Moreover, the non-automatic positioning of the tape in manual measurements is challenging, even when conducted by trained personnel experienced in measuring overweight and obese people [67]. This might apply to an even greater degree for larger waist circumferences among obese people, based also on the findings of other studies which report low reliability of manual WC measurements in obese subjects [34,68].

In our study (regarding model fit) WC performed better than BMI and WHtR in the prediction of the three different body fat measures (visceral, absolute, relative). Moreover, the association between the three manual anthropometric measurements with visceral fat mass was not linear, so we suggest not modeling this association with simple linear regression equations. In general, the univariable prediction of SMM performed lowest in our study. This is an indication that other body dimensions are more important in predicting muscle mass and that this prediction should be focused on the upper and lower extremities. Our findings support other studies which found only little differences between manually and scanned measures of waist circumference and their association with relative fat mass (as assessed by the same BIA device as in this study) [42]. While some studies have already shown a good validation of linear and circumferential measurements between 3D scans and manual anthropometry, there is not yet so much information on volume reconstructions. We provide here indications that these volumes have an added value, in our case especially in the estimation of skeletal muscle mass.

We furthermore support other studies that information gained from the scans results in high predictive values for body composition measures [41]. In our study we show that the multivariable prediction of body composition explains significantly more variability than the univariate prediction given by WC, BMI and WHtR. Thus, scanning the participants brings with it an additional benefit because within a short time a large number of additional body measurements are available, which markedly improves the prediction of body composition. Concerning the method of selecting individual variables from the whole catalogue of scanned measurements, there are not many other studies we can use for comparison. Previous studies have aggregated meta-measures to cluster body types [44,45] or have used deep learning [43]. Moreover, recent research has also focused on developing methods using the 3D geometry of the surface topography in order to predict body shape [51,69,70].

Strengths and limitations: One limitation of the present study is that body composition was estimated using BIA, which is not the gold standard. In a recent validation study [55], the Seca mBCA 515 BIA-device have been shown to be less reliable for visceral fat than other measures. However, in the setting of the presented study it was not feasible to perform invasive and more time-consuming examinations. We have only examined young Swiss men. However, the homogeneous sample also has advantages in that the precision of our results within the examination group is higher than that of a stratified sample. Also, our study was based on a relatively small number or subjects. However, since the subjects of the present study regarding BMI and WC are closely comparable to the total population of all conscripts in Switzerland (which covers >90% of a given male birth cohort) [71,72], we believe that our results are generalizable, at least for young men in Switzerland. A larger and more diverse group of subjects will be needed to produce results which are generalizable for a broader population. Also, information about migration background and ethnicity (which influences body shape) should be collected in similar studies, and statistical methods validated in this study have to be tested in other data sets. Last but not least, the geometry accuracy of the mesh surface reconstruction performed by the Anthroscan VITUSbodyscan software has not yet been validated [51,69,70], which might be relevant to the volume estimations in our study.

5. Conclusion

Digital anthropometry is currently transforming areas of clinical nutrition assessment and provides new research opportunities [73]. We provide evidence, that even in smaller and homogenous samples prediction of body composition can be improved by making use of a broad range of standard scan measurements via various regression techniques. However, in order to make the best possible use of this technology, further studies that continue to lay important ground work must follow. The use of a 3D full body scan proved to be feasible for population based studies, and to produce results that strongly correlate to manually measured anthropometric measures. Some systematic differences remain and need further investigation. However, the use of a 3D scan might help in solving difficult situations like manual WC measurements in very obese individuals. As WC showed to be the best indicator for body fat in our study, this fact is of relevance for epidemiological applications. The 3D body scanner also allowed the prediction of skeletal muscle mass in our study. If this result is confirmed in other studies, and especially in a more varied population, this might reduce the use of multiple devices in population studies in the future.

The 90 selected measurements, including names, system-ID, mean and Standard Deviation (SD).

Rho RFM indicates Spearman rank correlations with RFM, and Rho SSM correlations with SMM (only Rho>0.5 are reported to provide an overview, and Rho>0.8 are reported in bold numbers).

(DOCX)

Click here for additional data file.

Correlation (Spearman) matrix for the three anthropometric measurements (BMI, WC, WHtR).

(DOCX)

Click here for additional data file.

Correlation (Spearman) matrix for the four body composition measurements (visceral fat mass, AFM, RFM, SMM).

(DOCX)

Click here for additional data file.

(CSV)

Click here for additional data file.

2 Jan 2020

PONE-D-19-26892

Predicting body composition from 3D laser-based photonic body scans in young men

PLOS ONE

Dear Dr. Staub,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

The manuscript is well assessed by the two Reviewers; however, several major critiques are raised in the present form. Read carefully the Reviewers' comments and respond them appropriately.

We would appreciate receiving your revised manuscript by Feb 16 2020 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Masaki Mogi

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements:

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at http://www.plosone.org/attachments/PLOSOne_formatting_sample_main_body.pdf and http://www.plosone.org/attachments/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. Please see the following video for instructions on linking an ORCID iD to your Editorial Manager account: https://www.youtube.com/watch?v=_xcclfuvtxQ

3. Please remove your figures from within your manuscript file, leaving only the individual TIFF/EPS image files, uploaded separately.  These will be automatically included in the reviewers’ PDF.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: I Don't Know

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The authors presents a paper comparing both manual anthropometric measurements and measurements derived from a 3D laser photonic scanner to BIA measurements. The paper is well written but overly complex and the statistics needs to be slightly revisited.

Main considerations:

(1) The model construction AND the validation is based on the same dataset, so the validation comparing measurements derived from a 3D laser photonic scanner to BIA measurements will of course give a high comparability. Since the model was optimized for exactly these datasets. In order to complete a more reliable validation the model parameters should not be selected in the same group as the BA plots and regression testing was performed in. Sugget to either split the group in two where the model parameters are determined in the first sub-group and the validation is performed in the second sub-group, or perhaps the follow-up 54 datasets could be used for the validation?

(2) The research questions are overly complex. There are six distinct research questions identified in the aims and these are strongly related. Is it necessary with such complex and many research question? It seems, in the paper, as the main questions rather are (i) to define model parameters for fitting 3D laser photonic images to body composition, and (ii) to validate this model?

Minor consideration:

(1) In the discussion it is noted that participants were standing with their legs hip-width apart for the scanning, whereas they were standing with legs close together for the manual measurements. This will introduce a bias towards lower heights in the scanning. I presume the length of the legs could have been determined from the scanning parameters and therefor (simply using the Pythagorean theorem) the heights could have been corrected for?

Reviewer #2: This paper presents a study of using 3D digital measurements to predict body composition. However, the major issues for this paper are:

1) Vary lack of novelty. There are many existing papers conducted a similar study and this paper has no outstanding points comparing to the others.

2) Minimal contribution to the research in this area. The significance of this paper is unknown.

3) The experiment is not-well designed, using BIA as the ground truth is questionable. And the accuracy of the 3D scanner is not fully validated.

4) The dataset is too small and the diversity is limited. It is hard to guarantee the result can be generalized well.

A) Data Source

Dataset is very small. Age range and ethnicity diversity are limited. Statistics detail is not provided. And the justification for using such a small dataset is inadequate.

Using BIA as the ground truth is questionable. The accuracy of BIA is pretty low. The author does not give enough justification for using BIA as the ground truth.

B) Method

The author evaluated the agreement between manual and scanned anthropometric standard measurements. But the anthropometric is only limited to BMI, WC, and WHtR.

The geometry accuracy and measurement readability of the 3D scanner is not evaluated and justified, especially when users take a 3D scan with their T-shirts on. More evidence should be provided for the data cleanness.

How the feature selection is conducted (training set size, validation set size, and test set size) is not clear. I suggest the author use a separate dataset to test the performance of selected features.

The author states there is no overfitting but there is not enough justification (like experiments, data) for this.

The author states they collected 89 measurements, but does not provide enough detail about the 89 measurements. What are the measurements, how are they correlated, is there a better way to formulate the prediction function? Why only several of them are useful?

Data analysis is performed but the author did not dig into why some features are selected and why the others are irrelevant and what is the interrelation between the features. Other than feature extracted directly from the software, feature mapping may be explored to see if there are better features derived from the simple features.

The result lacks significance.

C) Other

Related work/background does not cover the latest research in body composition analysis using 3D geometry.

The manuscript is hard to read and bad-organized.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: Yes: Janne West

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

2 Apr 2020

A) Editor

“Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

The manuscript is well assessed by the two Reviewers; however, several major critiques are raised in the present form. Read carefully the Reviewers' comments and respond them appropriately.”

- Answer: Many thanks for giving us the opportunity to revise our manuscript. We are happy to comment on all points that the reviewers have noted. Please find our answers below.

B) Reviewer #1

“The authors presents a paper comparing both manual anthropometric measurements and measure-ments derived from a 3D laser photonic scanner to BIA measurements. The paper is well written but overly complex and the statistics needs to be slightly revisited.

Main considerations:

(1) The model construction AND the validation is based on the same dataset, so the validation compar-ing measurements derived from a 3D laser photonic scanner to BIA measurements will of course give a high comparability. Since the model was optimized for exactly these datasets. In order to complete a more reliable validation the model parameters should not be selected in the same group as the BA plots and regression testing was performed in. Sugget to either split the group in two where the model param-eters are determined in the first sub-group and the validation is performed in the second sub-group, or perhaps the follow-up 54 datasets could be used for the validation?”

- Answer: Thank you for this comment. As we wrote in the methods section of our article, we as-sessed the predictive value of our models using the standard method of leave-one-out cross-validation. The main reason for choosing this method was that our sample is relatively small. We have expanded the section on methodology to include this justification. The second reason was that our paper focuses on the potential benefit of including multiple predictors in a model, on the number of potentially relevant predictors and the way to select them, rather than the con-struction and testing of a single optimal predictive model. We used bootstrapping to assess the stability of predictor selection and showed that the identity of the predictors included in models varied substantially among bootstrap samples. We modified the title of the paper to clarify this purpose. The presentation of a single model with regression coefficients in Table 3 (multivari-ate lasso) may have suggested that this was proposed as optimal predictive model. In fact, Ta-ble 3 had a different purpose, which is why we did not validate this model. We agree that this was not sufficiently clear. To avoid increasing the complexity of the paper, we removed Table 3 and instead improved the description of individual predictors in the Supplementary Table 1. Finally, please note that the first part of the results directly compares manual and scanned measurement (BA and scatter plots) and does not involve any model construction or model pre-dictions. The measurements were chosen a-priori due to their established use in medical re-search and public health assessments. This is why no model validation was performed here.

“(2) The research questions are overly complex. There are six distinct research questions identified in the aims and these are strongly related. Is it necessary with such complex and many research question? It seems, in the paper, as the main questions rather are (i) to define model parameters for fitting 3D laser photonic images to body composition, and (ii) to validate this model?”

- Answer: We agree that the research questions were complex in the initially submitted manu-script version. We therefore reduced the number of research questions from six to three in the revised version. However, as explained above, our aim was not to define and validate a single model for the prediction of body composition, but rather to compare models based on two data sources (manual and scanner) as well as different approaches to predictor selection. Thus, we could not reduce the research questions as much as proposed above.

“Minor consideration:

(1) In the discussion it is noted that participants were standing with their legs hip-width apart for the scanning, whereas they were standing with legs close together for the manual measurements. This will introduce a bias towards lower heights in the scanning. I presume the length of the legs could have been determined from the scanning parameters and therefor (simply using the Pythagorean theorem) the heights could have been corrected for?”

- Answer: Thank you for this excellent suggestion. In fact, we addressed the systematic height dif-ference in a separate paper, which was just published in PeerJ (https://peerj.com/articles/8095/). In this sub-analysis we compared two scan positions in ca. 50 follow-up individuals (standard position with legs hip-wide apart vs. legs together and straight posture as being manually measured by an anthropometer). Our conclusion is that leg position is only partially responsible for the systematic height differences, the rest might be explained by additional factors such as slightly cut point clouds on the top of the head and at the bottom of the feet, as well as posture differences. As the PeerJ study has been published in the meantime, we complemented the discussion of the present study accordingly.

C) Reviewer #2

“This paper presents a study of using 3D digital measurements to predict body composition. However, the major issues for this paper are:

1) Vary lack of novelty. There are many existing papers conducted a similar study and this paper has no outstanding points comparing to the others.”

2) Minimal contribution to the research in this area. The significance of this paper is unknown.”

- Answer: Thank you for this comment. Indeed, the number of studies in the medi-cal/epidemiological 3D full body scan area has increased pleasingly in recent years. Neverthe-less, in our view, there is still a research gap in which methodologies can be used to extract the most and relevant information from the cloud of points in order to explore associations with disease risks etc. By following a new path (comparing different statistical techniques to select parameters) and by showing that even in a small sample we can improve the prediction of body composition using these techniques and the full battery of standard measurements, we believe that we are making a valuable contribution to the 3D full body scan community. We added a sentence highlighting our contribution to the field to the conclusion paragraph.

“3) The experiment is not-well designed, using BIA as the ground truth is questionable. And the accu-racy of the 3D scanner is not fully validated.”

- Answer: We admit that the paper's title may have suggested that we regard BIA measurements as "true" measures of body composition. This simplification was necessary to avoid an overly long title. We clearly state in the introduction and again in the limitations section that BIA is not the gold standard for determining body composition. As we also state in the limitations, BIA was simply the only applicable method in our study setting (limited time window available with-in the Armed Forces basic training and no possibility to use more invasive techniques). We would also like to note that even when still not interchangeable with DEXA the latest generation of BIA devices (as for example the SECA mBCA 515 used in our study) has made progress, as shown by independent validation studies (https://www.mdpi.com/2072-6643/10/10/1469). We pick up your point by adding at various places in the manuscript that our BIA measurements are an “estimate” of body composition. Regarding the accuracy of 3D scanners, our own and other studies have shown a high reliability and validity of linear measurements (lengths, cir-cumferences, etc.) in comparison to manual measurements (the repeatability is even increased against manual measurements due to the reduced intra-observer errors). This is also the case for the latest highly precise and accurate Anthroscan VITUSbodyscan device used in our study and also in large medical cohort studies in Germany (NAKO, Life). However, regarding vol-umes there are not yet many validations studies. We add to this research gap by showing that especially volumes contribute to the estimation of SMM. We supplemented this added value to the discussion.

“4) The dataset is too small and the diversity is limited. It is hard to guarantee the result can be general-ized well.”

- Answer: We agree that our sample is relatively small and generalizability may be limited. We also emphasize this fact at various points in our manuscript. However, we believe that this is al-so one of the contributions of our study: We show with that even with a small sample from a homogeneous population, multiple measurements derived from the scanner contribute to the prediction of body composition. We also show (Table 2) that the identity of the selected predic-tors is variable. We believe that the statistical methods (parameter selection) that we point out are transferable and can provide other studies with valuable information. We agree that the presentation of a single model with regression coefficients in Table 3 (multivariate lasso) may have suggested that this was proposed as optimal predictive model. We fully agree that this model is specific to our study population and may be not valid for people of different age and ethnicity. As explained above (answer to reviewer 1), we removed Table 3 from the manuscript to avoid a misunderstanding on its purpose.

“A) Data Source

Dataset is very small. Age range and ethnicity diversity are limited. Statistics detail is not provided. And the justification for using such a small dataset is inadequate.”

- Answer: Thank you for this comment. For sample size and diversity, see our previous answer. Regarding statistical detail, we added some more information about age distribution and eth-nicity of our probands in the data description. Further information is provided in Appendix Table 1.

“Using BIA as the ground truth is questionable. The accuracy of BIA is pretty low. The author does not give enough justification for using BIA as the ground truth.”

- Answer: For general use of BIA, see our answer to your third comment above. We added a little more text in the methods section, why we had to use BIA.

B) Method

“The author evaluated the agreement between manual and scanned anthropometric standard measure-ments. But the anthropometric is only limited to BMI, WC, and WHtR.”

- Answer: Unfortunately, we only had a limited time window available for measurements within normal Armed Forces operations. We were therefore unable to collect any other manual an-thropometric mass than these standard measurements (BMI, WC, WHtR). And more time would indeed be required to take other high-quality measurements such as the circumferences of the thigh or forearm. However, we think that for most readers of our article, the comparison be-tween manual and scanned measurements is particularly interesting for the standard measure-ments we compare (BMI, WC, WHtR), which are often used in medical and epidemiological studies.

“The geometry accuracy and measurement readability of the 3D scanner is not evaluated and justified, especially when users take a 3D scan with their T-shirts on. More evidence should be provided for the data cleanness.”

- Answer: Thank you for this important note that we were not precise enough about our data preparing process. We worked with the raw point clouds for the extraction of all standard measurements (but not the volumes). For the standard measurements, there was no postpro-cessing of the scans, only all 104 scans were checked for their quality (artefacts, etc.). For the calculation of the partial volumes, we automatically calculated closed surfaces using the stand-ard procedure in the Anthroscan software (good quality level, medium mesh size), the cutting off of the partial volumes was also done fully automatically via the Anthroscan software, but was supervised by us. We added corresponding text to the methods section. Regarding geomet-ric accuracy (which mostly applies for postprocessed scans (mesh)): We agree that, even with the high-precision Anthroscan VITUSbodyscan devices, there has not yet been a study that has validated the accuracy of the overlying mesh. However, the resolution of these devices is so high (even tattoos are precisely recorded, and these devices are also used in dermatology) that we are confident about their accuracy. However, this is less relevant in the context of the present study because we automatically extract most of the measurements from the raw point cloud (over 7 million points) and no proband wore a T-shirt during the measurements. However, we will in-clude this important point when we work with 3D surface geometry in future studies. And, we have added a corresponding sentence to the limitations section.

“How the feature selection is conducted (training set size, validation set size, and test set size) is not clear. I suggest the author use a separate dataset to test the performance of selected features.”

- Answer: Thank you very much for this comment. Please see our answer to the equivalent first comment of Reviewer 1.

“The author states there is no overfitting but there is not enough justification (like experiments, data) for this.

The author states they collected 89 measurements, but does not provide enough detail about the 89 measurements. What are the measurements, how are they correlated, is there a better way to formulate the prediction function? Why only several of them are useful?”

- Answer: Regarding overfitting, we present evidence in Figure 3. In this visualization, curves il-lustrate how many predictors were needed to obtain the maximal predictive value and the mod-erate degree of overfitting (difference between r2 and cross-validated r2) found even with multi-ple predictors. We improved the corresponding part in the results section in order to better high-light the overfitting question. Regarding details on the measurements: We agree that some of these measurement designations are somewhat burdensome. But please note that we provide the exact system name and official ID numbers of all measurements used (based on standards ISO 7250 / ISO 8559 and DIN EN ISO 20685) together with descriptive statistics in Appendix Table 1. This should also facilitate comparison across studies with the same scanner device and soft-ware. Regarding correlations, we already noticed in our manuscript that some of these meas-urements are highly correlated. We added information on this to the main text as well as to supplementary Table 1, and we newly added Supplementary Figure 1. Regarding the prediction formula: As noted above, the aim of our paper was not to provide a single optimal prediction function, and we removed Table 3 to avoid this misunderstanding. We also state more clearly in the discussion that the optimal prediction function (possibly considering further derived fea-tures) still has to be determined.

“Data analysis is performed but the author did not dig into why some features are selected and why the others are irrelevant and what is the interrelation between the features. Other than feature extracted di-rectly from the software, feature mapping may be explored to see if there are better features derived from the simple features.”

- Answer: Thank you for this good comment. We illustrated the relationships among features in a new Supplementary figure (tree from cluster analysis). Furthermore, we classified the features into groups (height, girth, volume etc.) and evaluated correlations within and between groups, as well as their correlations with body composition (addition to Appendix Table 1). We could have made similar considerations for the multivariable models, but this was beyond the scope of the paper and would have made it even more complex. While other studies have attempted to make sense of the standard measurement batterie using PCA and other methods, our goal has been to work relatively close to the standard measures that the Anthroscan software provides. Since we have already used 90 measures for our predictions within 104 subjects, additional or combined measures would most probably caused overfitting issues.

“The result lacks significance.”

- Answer: Please see our answer to your first comment.

C) Other

“Related work/background does not cover the latest research in body composition analysis using 3D geometry.”

- Answer: We were happy to take up this interesting new direction in the discussion.

“The manuscript is hard to read and bad-organized.”

- Answer: Thank you for this comment. We are aware that our manuscript is relatively complex. We have reduced the number of research questions (based on a comment from Reviewer 1), but it is in the nature of these remaining questions that their answers are and remain technical and therefore complex. Nevertheless, we have tried to reduce the complexity here and there and to offer more reader guidance.

Submitted filename: Answers_R1.docx

Click here for additional data file.

21 Apr 2020

PONE-D-19-26892R1

Multiple measures derived from 3D photonic body scans improve predictions of fat and muscle mass in young Swiss men

PLOS ONE

Dear Dr. Staub,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

==============================

Major revisions are still needed in the present form. See the Reviewers' comments and respond them appropriately.

==============================

We would appreciate receiving your revised manuscript by Jun 05 2020 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Masaki Mogi

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: (No Response)

Reviewer #2: Partly

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: (No Response)

Reviewer #2: N/A

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: (No Response)

Reviewer #2: (No Response)

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: (No Response)

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: (No Response)

Reviewer #2: I have the same concern with the Reviewer #1 on how the model is evaluated. And I totally agree with Reviewer#1 that the author should separate the dataset, i.e., train the model using one part and do the evaluation using the other. At lease I think the author should give that result as part of the analysis.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: Yes: Janne West

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

22 Apr 2020

“I have the same concern with the Reviewer #1 on how the model is evaluated. And I totally agree with Reviewer#1 that the author should separate the dataset, i.e., train the model using one part and do the evaluation using the other. At lease I think the author should give that result as part of the analysis.”

Answer: Thank you very much for insisting on this really important point. We agree that models should be trained and tested on different data. We would like to stress again that this is exactly what we did when we applied leave-one-out cross-validation: We trained a model on (n-1) sub-jects and tested it on the remaining (excluded) subject, and repeated this in turn for all subjects in the sample. This procedure avoided the arbitrary subdivision of our rather small sample into even smaller subsamples. Please let us mention that Reviewer #1, who initially suggested the da-taset splitting, agreed with our explanation and with the use of this procedure for model evalua-tion given the aims of our study. We therefore believe that we have already sufficiently ad-dressed this Reviewer's concern and hope that we could clarify this here.

Submitted filename: Answers_R2.docx

Click here for additional data file.

29 May 2020

Multiple measures derived from 3D photonic body scans improve predictions of fat and muscle mass in young Swiss men

PONE-D-19-26892R2

Dear Dr. Staub,

We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements.

Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication.

Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at https://www.editorialmanager.com/pone/, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

With kind regards,

Masaki Mogi

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

The authors responded to the Reviewer's comments.

Reviewers' comments:

2 Jun 2020

PONE-D-19-26892R2

Multiple measures derived from 3D photonic body scans improve predictions of fat and muscle mass in young Swiss men

Dear Dr. Staub:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Masaki Mogi

Academic Editor

PLOS ONE