Automatic Segmentation and Quantification of Thigh Tissues in CT Images

Abstract

Quantification and distribution of the thigh adipose tissues in CT images have clinical implication in prognostic chronic disease including type 2 diabetes and osteoarthritis. Although there are studies in the literature addressing the quantification of thigh tissues, only a handful of them aims to segment and quantify thigh adipose tissues in CT images automatically. In this study, we propose an automated technique for the segmentation and quantification of muscle, inter- and intra-muscular adipose tissue and subcutaneous adipose tissue in thigh CT images. Our technique combines morphological operations, thresholding, a Gaussian mixture model and the use of an accumulator matrix to map the number of adipose tissue pixels about muscle pixels and thus, to allow an automatic differentiation between SAT and Inter-MAT. Our method was quantitatively assessed using 144 thigh images extracted from 72 leg (left and right) CT scans. All images were manually segmented and the tissues quantified by a specialist with the help of a computer software and used for further comparative analysis. Our technique obtained precision of 0.998 and 0.982, respectively, for the fascia and thigh regions with corresponding recall values of 0.978 and 0.975. Also, the Dice similarity coefficient for both areas was close to 0.98.