Effect of Single-Slice CT Segmentation Methods on Fat Volume and Body Shape Estimation

Abstract

Several automatic and semi-automatic algorithms for adipose tissue (AT) segmentation in CT have been proposed. Our study aimed to determine the effect of the preselected HU range, on the resulting AT volumes and establish whether there is a relationship between body shape and AT values determined, from a single slice. Scans of 98 patients acquired using two CT protocols, were used. Three axial slices were selected from each subject's CT data. Subcutaneous and visceral adipose tissues (SAT, VAT) were manually segmented and analyzed using three different HU ranges. In addition, a simple BMI calculation model was created by the segmented data. The areas segmented with the three different HU ranges, correlated well with each other in the case of SAT (r2=0.99, and r2=0.99) and VAT (r2=0.99, and r2=0.998). The preselected slice position had no significant effect on correlation; however, the absolute values of ATs were statistically different. CT image data acquired with higher tube current yielded a better correlation between SAT, VAT, and BMI. We also found that the correlation of VAT area to mass and BMI was weaker than the corresponding SAT correlations. The simple model-based BMI estimation is in line with real BMI data (males: r2=0.78, females: r2=0.841). The segmentation threshold does not substantially affect correlation, among the segmented AT values; however, their absolute values are significantly different. In addition, and interestingly, the body shape can be accurately described from the segmented AT data from a single CT slice.