In this sample of healthy adults aged 18-77 years of age, estimates of visceral adipose tissue volume using the recently-introduced Hippo Fat™ program were highly correlated with estimates using slice-O-matic™, a semi-automated image segmentation program widely used to measure visceral adiposity (e.g., Janssen et al. 2002
; Janssen et al. 2002
; Lee et al. 2004
; Potretzke et al. 2004
; Shen et al. 2004
; Shen et al. 2004
) that has been validated against directly measured chemical composition (Mitsiopoulos et al. 1998
). Although there was good overall agreement between the methods (R2
>0.9 and slope of nearly 1.0), they produced significantly different estimates of SAT and VAT at both the group and the individual levels. Average VAT was underestimated by 9.4% and average SAT was overestimated by 3.7% using Hippo Fat™ compared to slice-Omatic™. The overestimation of SAT tended to be greater among individuals with higher SAT levels. Converting the volumetric units to mass, the inter-method difference in SAT was 0.18 kg and the difference in VAT was 0.38 kg. Individual-level error for VAT was also substantial; Hippo Fat™ gave estimates of VAT ranging from 1.07 kg less to 0.51 kg more than estimates from the same person using slice-O-matic™. Therefore, slice-O-matic™ and HippoFat™ do not provide interchangeable results for total abdominal VAT or SAT volume. When designing a clinical or epidemiological study, the same image analysis method should be used for all observations and at all study centers. This would be particularly important for longitudinal and intervention studies where a change in image analysis software from baseline to follow-up could substantially bias the results for or against an effect on visceral adiposity.
While we did not examine the agreement between the methods on a slice-by-slice basis, it is possible that agreement may be greater or lesser at specific individual slices or when examining particular adipose tissue depots (e.g., mediastinal, pericardial, retroperitoneal). We found, for example, that for the image taken at the L4-L5 intervertebral space, the single image location most commonly used for MRI studies of visceral adiposity, the degree of overestimation of SAT area and the degree of underestimation of VAT area were lower than we found for total SAT and VAT volumes.
Hippo Fat™ software can be run without user supervision or manual editing, and can process 32 images from a single subject in approximately six minutes. This compares very favorably to the 60 minutes required to process the same number of images using slice-O-matic™. However, when we first attempted to use Hippo Fat™ without performing manual editing, the inter-method difference was ~19% for VAT and ~10% for SAT (data not presented). We therefore took advantage of the manual editing features of Hippo Fat™ to adjust both the VAT contour and the Gaussian distribution settings, which increased the processing time to 30 minutes per individual having a set of ~28 MR images. In the case of individuals with low levels of adipose tissue, the non-adipose tissue and the adipose tissue peaks in the histogram somewhat overlapped one another, creating the need to edit the Gaussian curves manually. This issue has been previously noted (Positano et al. 2004
Even in the case of heavier subjects with large amounts of visceral adipose tissue, the automated contours for the visceral compartment appeared to exclude some of the adipose tissue from the subsequent analysis, perhaps because of varying intensity of the VAT signal across a single image, or poor resolution of the boundary between the abdominal wall and the visceral compartment. If the signal intensity for VAT is low in a region, the automated VAT contour may not detect it, and the VAT level may be underestimated. Image quality may therefore determine if Hippo Fat™ can be used successfully in its completely automated mode, or if manual adjustments will be required. Subjects in the present analysis were imaged on a 1.5 T scanner using a standard body coil with a T-1 weighted fast spin-echo pulse sequence — a fairly standard protocol for the acquisition of abdominal VAT data in existing studies and similar to the protocol used to initially test the Hippo Fat™ method (Positano et al. 2004
). Better resolution images may improve the ability of Hippo Fat™ to generate accurate estimates of VAT without manual editing.
A disadvantage of the Hippo Fat™ program is that it is designed to estimate the total amount of SAT and VAT in the abdominal region only. Recent work has established that there are sub-compartments of adipose tissue within both the subcutaneous compartment, including deep and superficial adipose tissue (Kelley et al. 2000
; Smith et al. 2001
) as well as the visceral adipose tissue compartment, including pericardial adipose tissue, mediastinal adipose tissue, and others (Iacobellis et al. 2003
) that may play particularly important roles in the development of disease (Shen et al. 2003
). Computer-assisted manual editing programs such as slice-O-matic™ allow particular adipose tissue sub-compartments to be visually identified and tagged and therefore hold a key advantage for flexible body composition assessment.
In conclusion, there was good overall agreement between Hippo Fat™, a newly introduced automated method for measuring visceral adipose tissue in humans, and a widely used, validated semi-automated software. However, the two methods produced significantly different estimates of SAT and VAT at both the group and the individual levels and are therefore not interchangeable. When designing a clinical or epidemiological study, the same image analysis method should be used for all observations and at all study centers. Advantages of Hippo Fat™ are low cost and a significant reduction in the time required for image analyses, which becomes particularly important in studies where large numbers of participants and/or large numbers of images per participant must be analyzed. Disadvantages of Hippo Fat™ are that the program does not allow for measurement of sub-compartments within the subcutaneous and visceral adipose tissue depots and that manual editing may be required, unless the MR images being used have excellent resolution and quality.