Preventing overestimation of pixels in computed tomography assessment of visceral fat

Computed tomography (CT) is a common research procedure for measuring abdominal fat distribution, but little is written about the software used to analyze images. Our objective was to compare in-house and commercially available software for quantitative measurement of abdominal fat distribution. In the process, we encountered some unexpected problems.

A total of 123 volunteers had single-slice abdominal CT images taken that were used to evaluate various aspects of the commercial image analysis program.

The agreement between the commercial and in-house programs was excellent (r = 0.996, p < 0.00,001) for both total and intraabdominal fat, and we were able to reduce between-observer variability in measured fat areas through the use of statistical handling of region of interest information. We also noted that intracolonic contents sometimes had the same Hounsfield units as adipose tissue. We analyzed single-slice CT images from 50 volunteers to determine the potential impact of this effect on visceral fat area; the overestimate of visceral fat area was 19 +/- 22% (maximum, 112% overestimate). The commercial program could prevent this error, whereas our in-house program could not.

We concluded that a readily available commercial image analysis program compares well with a previously validated in-house program and that it offers some advantages with respect to preventing overestimation of pixels as visceral fat.