Evaluation of Low Bone Mineral Mass Using a Combination of Peripheral Bone Mineral Density and Total Body Composition Variables by Neural Network

Abstract

The aim of this work was to evaluate low bone mass using the feed-forward neural network (NN) with good accuracy taking into account the forearm and heel bone mineral density (BMD) as well as total body composition variables. A total number of 162 subjects including 88 women (mean ± SD age = 37.7 ± 15.2 years) and 74 men (mean ± SD age=31.3 ± 10.9 years) were studied. In each subject, BMD (g cm-2) at forearm and heel using peripheral dual-energy X-ray absorptiometry (pDXA) and total body composition variables by multifrequency bioelectrical impedance analyzer were measured. The measured forearm BMD was used to estimate femur neck BMD by DXA using the published formula. Based on its T-score value, subjects were classified as normal and low bone mineral mass groups separately. In women, it was found that the forearm BMD was positively correlated with body fat percentage (r=0.327; p<0.001). It was observed that 27% of women and 15% of men were affected by low bone mass. In the NN modelling, the following 10 measured variables were used in men and women separately: i) BMI ((kg/m2); ii) average forearm BMD (g/cm2); iii) average heel BMD (g/cm2); iv) body fat (%); v) muscle mass (kg); vi) visceral fat index; vii) bone mineral mass (kg); viii) total body water, TBW (%); ix) basal metabolic rate, BMR (kCal); and x) metabolic age (years). Analysis of low bone mineral mass evaluation using NN projected an accuracy of 87.5% and 85.1% in women and men population, respectively. With the aid of BMD at peripheral skeletal sites and total body composition variables, low bone mass can be evaluated with good accuracy.