Accurate and reliable serial measurement of body composition over the early life period is critical for understanding the etiology and progression of childhood obesity. However, current approaches are limited in their ability to provide safe and low-cost measures of sufficient reliability in pediatric populations. Here, we validate Diffuse Optical Spectroscopic Imaging (DOSI), a fast, non-invasive low-cost imaging tool with no known safety risks, to quantify body fat% (BF%) in early life.


DOSI was used in N=35 children between 3-6 yrs age to derive measures of fat fraction (DOSI-FF) sampled at a source-detector separation of 28mm using six different locations [bicep (BS), tricep, subscapular (SS), calf, central abdomen (CA), lower back]. A linear regression model was used to associate DXA-based whole body fat percentage (DXA-BF%) with the six-site DOSI-FF measures. Following this, a best subsets analysis was used to regularize the model and avoid overfitting for out-of-sample prediction. The best subsets defined (CA only) model was used to predict DXA-BF% and compared with a simple “clinical” model based on age, sex and BMI.


The six-site regression model was significantly associated with DXA-BF% (R^2=82%; three sites p<0.05: CA, BS, SS). Best subsets analysis converged on a single measurement: CA (BIC=-34). Single-site (CA) out-of-sample DOSI-based BF% was strongly predictive of DXA-based BF% (R2 =76%, p<10e-10). In comparison, the out-of-sample “clinical” model was only moderately predictive of DXA-based BF% (R^2 =40%, p<10e-5).


DOSI represents a novel tool that is feasible and safe for percent body fat measurement in a very young pediatric population, such that, DOSI-based BF% strongly and significantly predicts DXA-based BF%.