MRI adipose tissue segmentation and quantification in R (RAdipoSeg)

Abstract

Background: Excess adipose tissue is associated with increased cardiovascular and metabolic risk, but the volume of visceral and subcutaneous adipose tissue poses different metabolic risks. MRI with fat suppression can be used to accurately quantify adipose depots. We have developed a new semi-automatic method, RAdipoSeg, for MRI adipose tissue segmentation and quantification in the free and open source statistical software R.

Methods: MRI images were obtained from wild-type mice on high- or low-fat diet, and from 20 human subjects without clinical signs of metabolic dysfunction. For each mouse and human subject, respectively, 10 images were segmented with RAdipoSeg and with the commercially available software SliceOmatic. Jaccard difference, relative volume difference and Spearman’s rank correlation coefficients were calculated for each group. Agreement between the two methods were analysed with Bland–Altman plots.

Results: RAdipoSeg performed similarly to the commercial software. The mean Jaccard differences were 10–29% and the relative volume differences were below ( ±) 20%. Spearman’s rank correlation coefficient gave p-values below 0.05 for both mouse and human images. The Bland–Altman plots indicated some systematic and proporitional bias, which can be countered by the flexible nature of the method.

Conclusion: RAdipoSeg is a reliable and low cost method for fat segmentation in studies of mice and humans.