Abstract
Quantification of distinct subcutaneous and visceral fat regions in the abdomen is essential in clinical studies of metabolic disorders and cardiovascular disease. Computed Tomography (CT) is a widely adopted imaging technology for abdominal fat quantification because the intensity range of fat in Hounsfield Units (HU) is distinct from other tissues in the pelvis and abdomen. Nevertheless, it has been observed that the quantification of visceral fat based solely on intensity is subject to errors caused by food residues in the intestines that may have intensities similar to fat. Herein we present a method for automated quantification of abdominal fat in CT with emphasis on reducing errors in visceral fat measurements caused by food residues. The fat pixels are first identified in the feature space of HUs and then divided into subcutaneous and visceral component using anatomic location. Food residues within the intestines that are previously inaccurately labeled as visceral fat (false positives) are identified and removed using a machine learning technique. Experimental results include validation against reference data over 144 CT images to test the generalization capability of our scheme.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have