Scope of the study
To test the pertinence of the side view in body scales, (i) front and side photographs were taken and we simultaneously collected a series of bio-anthropometric measurements in an African sample (161 subjects). For both front and side photographs, (ii) we first digitized the silhouettes and quantified the variation in body shape of these silhouettes using a geometrical method (Normalized Elliptic Fourier Descriptors, or NEFD). The computation of NEFD allowed the characterisation of the shape of silhouettes by objective numeric values. Finally, (iii) we analysed how body shape as measured by NEFD covaried with various bio-anthropometric indices, and identified which body view (side or front) best reflected health and nutritional status (as defined by our indices).
We selected two large ethno-linguistic groups: Western Bantoid (Sahel region) and Bantu (equatorial forest region) which present genetic  and linguistic differences , and belonging to the language area of Africa covering the major part of sub-Saharan Africa (mainly Western, Central and Southern Africa), the Niger-Congo. These two groups differ in their macroscopic phenotypes, the first being slender and the latter more stocky . Our sample therefore captured a continuum of weight gain for contrasting body morphologies and thereby represents a wide range of body shape variation among African populations. Specifically, we sampled two populations native to Western Bantoid and Bantu regions: 81 Senegalese from Dakar (31 males and 50 females) and 80 Cameroonians from Yaoundé (51 males and 29 females) during a 2-month visit in 2009.
Producing the photographs
Subjects wore close-fitting black sportswear and were photographed after removing jewellery/distinctive objects. One front and one side (left) photograph was taken for each individual, after checking that they were standing straight with feet at 50 cm apart, arms making an angle of approximately 45° with the trunk and with palms of their open hands facing forward. All photographs were taken using a Canon 450D camera mounted on a tripod 1 m above the ground and 3.6 m away from the subject with the same 35 mm zoom (equivalent to a normal lens on a full frame camera).
To define the morphology of sampled individuals, the same researcher (E.C.) took a set of anthropometric measurements that were then condensed into 9 anthropometric indices. Body Mass Index (BMI, #1) was calculated by dividing body mass (in kg) by the height squared (in m2). The percentage of body fat (hereafter, fatness, #2) was derived from: sum of biceps, triceps, supra-iliac and subscapular skin folds . The somatotype profile is a ternary composite expression of slimness index (i.e. ectomorphy, #3), musculature index (i.e. mesomorphy, #4) and fatness index (i.e. endomorphy, #5), used to characterize the morphology of subjects . Body fat distribution was assessed by two proxies: waist circumference (WC, #6) and waist to hip ratio (WHR, #7), as specific measurements of abdominal obesity. Furthermore, mean blood pressure (mean BP, #8) was defined as: (diastolic BP + 1/3) x (systolic BP-diastolic BP), and a measurement of fasting blood glucose (glycaemia, #9) was taken with a glucometer. These last two indices have been shown to be associated with body weight , and identifying whether these physiological variables are associated with a specific view of human shape is relevant in the context of the use of body scales in the prevention of the valorisation of overweight as a risk factor for obesity [5–7].
All bio-anthropometrics were obtained following standardized procedures. Height was measured to the nearest millimetre using a portable stadiometer (SiberHegner, Switzerland). Body weight was measured in very light clothing, to the nearest 100 g, using a digital beam scale (Tanita, Japan). Computing the somatotypes was based on height and weight, tricipital, subscapular, suprasinal and medial calf skinfolds, biepicondylar humerus and femur, and arm (flexed and tensed) and calf circumferences. Circumferences were measured to the nearest millimetre, in a standing position using a non-stretchable tape measure. Skinfolds were measured using a Harpenden skinfold caliper (Holtain Ltd. UK). The biepicondylar humerus and femur bone breadths were measured using a Mitutoyodial calliper. WC was measured mid-way between the lowest rib and the iliac crest, at the end of a gentle expiration. Hip circumference was measured at the greater trochanters. Mean BP was derived for the average of two diastolic and systolic blood pressure (BP) readings, taken with the subject in a seated position, after a 15 minutes rest.
BMI, fatness, and the ternary expression of the somatotype were highly correlated (Spearman correlations, absolute rho = 0.41-0.96 for men; absolute rho = 0.64-0.95 for women). Therefore, we performed Principal Component Analysis (PCA) on these metrics (unscaled and centred) for each sex, extracting the first principal component (PC) which captured 87 % and 89 % of the total variation in men and women, respectively. This PC was then used as a synthetic measure of the overall body size defined as corpulence, our bio-anthropometric #10. PCA as well as all other analyses were performed, unless stated otherwise, using the R (version 3.1) statistical software.
Body shape analysis
To assess and visualize the influence of all 10 bio-anthropometric indices on body shape, we used the elliptic Fourier shape analysis , which has been shown to be reliable for the study of variation in human body shape . Briefly, during this analysis the outline of the body is expressed as x and y Cartesian coordinates and characterised by a function of the curvilinear abscissa representing the net distance on the outline from an arbitrary starting point [18, 28]. As such, the body outline corresponds to a periodic signal which can be approximated by a sum of trigonometric functions following traditional Fourier series expansions.
To conduct this Fourier analysis, we first delimited body outlines for each original photograph using a virtual paintbrush using Adobe Photoshop CS. We then removed hands and hair from outlines in order to suppress variation that was outside the scope of the present study. We also removed arms from side views when they overlapped with the body outline. All photographs were then transformed into the portable anymap file format (*.pnm) and loaded in R using functions from the pixmap package . Outlines obtained were digitised and we computed NEFD associated with each outline . NEFD are a set of coefficients that parametrise the trigonometric functions used to approximate body outlines. For all pictures, we consider a number of Fourier harmonics corresponding to a perfect pixel approximation for the shortest outline (between 3208 and 4648 depending on the view orientation and sex).
Second, we studied the influence of all 10 bio-anthropometrics on body shape, by reducing the dimensionality of the shape information by applying a PCA on the NEFD obtained for each combination of sex and view (discarding the three coefficients of the first harmonic that are constrained by normalization) . We retained the first 8 principal components from each PCA. Together, they captured around 95 % of the total body shape variation for front and 99 % for side outlines for each sex. Then, we performed a multivariate regression for each bio-anthropometric index, sex and view (Additional file 1: Table S1, Additional file 2: Table S2, Additional file 3: Table S3 and Additional file 4: Table S4). In all regressions, the 8 PCs associated with NEFD defined the multivariate response and the independent variables considered were: the bio-anthropometric, height and population. For the analysis of front outlines, we also considered the angles of arms and legs (measured using imageJ software) as covariates so that variation in body shape caused by posture differences did not impede analysis.
In order to quantify the extent to which the bio-anthropometrics influence body shape, we measured the coefficient of variation in perimeter-area ratio (PAR) associated with a change in four standard deviations (mean + 2SD vs. mean-2SD) for each bio-anthropometric index. The PAR is a simple metric proposed to relate to body image perception and known to strongly correlate with different body shape measures . For example, Courtiol et al.  showed that the PAR is strongly negatively correlated to BMI in a sample mainly constituted of Caucasians. To measure this metric, we predicted the corresponding NEFD values using the regression models refitted on all NEFD. We then reconstructed the outlines corresponding to these predictions and measured their PAR.
The study was approved by the Institutional Ethics Committee of the Institute of Medical Research and Medicinal Plant Studies of Cameroon. Oral consent was obtained from subjects, after they were fully informed about the study goals and methods.