Automated Fat Measurement and Segmentation with Intensity Inhomogeneity Correction

摘要

Adipose tissue (AT) content, especially visceral AT (VAT), is an important indicator for risks of many disorders, including heart disease and diabetes. Fat measurement by traditional means is often inaccurate and cannot separate subcutaneous and visceral fat. MRI offers a medium to obtain accurate measurements and segmentation between subcutaneous and visceral fat. We present an approach to automatically label the voxels associated with adipose tissue and segment them between subcutaneous and visceral. Our method uses non-parametric non-uniform intensity normalization (N3) to correct for image artifacts and inhomogeneities, fuzzy c-means to cluster AT regions and active contour models to separate SAT and VAT. Our algorithm has four stages: body masking, preprocessing, SAT and VAT separation, and tissue classification and quantification. The method was validated against a manual method performed by two observers, which used thresholds and manual contours to separate SAT and VAT. We measured 25 patients, 22 of which were included in the final analysis and the other three had too much artifact for automated processing. For SAT and total AT, differences between manual and automatic measurements were comparable to manual inter-observer differences. VAT measurements showed more variance in the automated method, likely due to inaccurate contours.