Automated quantification of body fat distribution on volumetric computed tomography

To develop a computerized method to automatically quantify visceral and subcutaneous fat distribution within the abdomen and pelvis on volumetric computed tomographic (CT) images.

Given the slices of interest, the algorithm automatically delineates a contour that separates the visceral fat from the subcutaneous fat on each slice. Explicitly, starting with extraction of the body perimeter, radii at a fixed angle increment are drawn from the perimeter to the center of the body. Along each radius, intensity profile is analyzed to determine the point on the subcutaneous fat layer that is closest to the body center (inner point). All inner points are then connected to form an inner contour, and a specific smoothing algorithm is subsequently applied to correct suboptimal results. Pixels having HU values between -190 and -30 are considered fat pixels. This procedure is repeated on each of the slices of interest. The visceral and subcutaneous fat volumes computed automatically were compared with those after the radiologist's adjustments. Ratios of volumetric visceral fat-to-total fat and visceral fat-to-subcutaneous fat were compared on average and with single-slice measurements obtained at L4 and L5 vertebral body levels.

Subcutaneous and visceral fat were automatically segmented using this algorithm on 419 axial CT slices in 9 CT scans (patients) within the abdomen and pelvis. The overall average percentage difference between the automated segmentation and the segmentation edited by the radiologist were 1.54% for the visceral fat and 0.65% for the subcutaneous fat.

Preliminary results have shown that total compartmental fat, including visceral and subcutaneous fat, can be automatically and accurately segmented on volumetric CT.