Classifying Abdominal Fat Distribution Patterns by Using Body Measurement Data

Abstract

This study aims to explore new categorization that characterizes the distribution clusters of visceral and subcutaneous adipose tissues (VAT and SAT) measured by magnetic resonance imaging (MRI), to analyze the relationship between the VAT-SAT distribution patterns and the novel body shape descriptors (BSDs), and to develop a classifier to predict the fat distribution clusters using the BSDs. In the study, 66 male and 54 female participants were scanned by MRI and a stereovision body imaging (SBI) to measure participants’ abdominal VAT and SAT volumes and the BSDs. A fuzzy c-means algorithm was used to form the inherent grouping clusters of abdominal fat distributions. A support-vector-machine (SVM) classifier, with an embedded feature selection scheme, was employed to determine an optimal subset of the BSDs for predicting internal fat distributions. A five-fold cross-validation procedure was used to prevent over-fitting in the classification. The classification results of the BSDs were compared with those of the traditional anthropometric measurements and the Dual Energy X-Ray Absorptiometry (DXA) measurements. Four clusters were identified for abdominal fat distributions: (1) low VAT and SAT, (2) elevated VAT and SAT, (3) higher SAT, and (4) higher VAT. The cross-validation accuracies of the traditional anthropometric, DXA and BSD measurements were 85.0%, 87.5% and 90%, respectively. Compared to the traditional anthropometric and DXA measurements, the BSDs appeared to be effective and efficient in predicting abdominal fat distributions.