Accurate quantification of visceral adipose tissue (VAT) using water-saturation MRI and computer segmentation: Preliminary results




To describe and evaluate the accuracy of water-saturation MRI and a computer segmentation program for quantification of visceral adipose tissue (VAT).

Materials and Methods

MRI was performed on five patients with whole-volume coverage of the abdomen using two different sequences: 1) a T1-weighted spoiled gradient-echo breath-hold sequence (non-water-saturation) and 2) a T1-weighted spoiled gradient-echo water-saturation breath-hold sequence (water-saturation). The computer segmentation program analyzed the data and calculated VAT volumes (cm3) from both sequences. The data from one patient were additionally processed with the use of a manual technique. The intrastudy reproducibility of the proposed method using the water-saturation MRI sequence and the computer segmentation technique was tested by repeated measures of the automated system analysis (×10) on MRI data from a single subject to calculate variability.


VAT volumes measured by the water-saturation MRI sequences were consistently greater than those measured by the non-water-saturation sequences. Comparison of VAT volumes derived from the water-saturation images and measured by the computer segmentation technique vs. the manual technique showed good correlation (K = 0.8), with a significant time-saving benefit associated with the automated method (5 minutes vs. 1 hour). There was poor correlation between VAT volume measurement calculated by the manual technique and the computer segmentation technique using non-water-saturation images. The reproducibility of the computer segmentation technique using data derived from water-saturation images was high, with a low variability (±5%).


The results obtained demonstrate that the proposed method may be able to provide accurate quantification of VAT in a highly reproducible and efficient manner. J. Magn. Reson. Imaging 2006. © 2006 Wiley-Liss, Inc.

OBESITY AND PHYSICAL INACTIVITY are global epidemics that warrant the immediate attention of the health-care community. An estimated two-thirds of Americans are overweight or obese (1). Current data show that each year in the United States alone, obesity leads to medical costs of over $90 billion and over 300,000 premature deaths (2). Following closely behind tobacco smoking, obesity and sedentary lifestyle are the leading causes of preventable mortality in the country (3). Excess fat in the central (visceral/abdominal) vs. the peripheral (subcutaneous/guluteofemoral) part of the body, independently of overall obesity, is closely linked to many facets of metabolic syndrome, including glucose intolerance, hypertension, dyslipidemia, and insulin resistance (IR) (4–6). Individuals with a high accumulation of visceral abdominal fat have an increased risk for developing type 2 diabetes and cardiovascular disease (7–9). The ability to accurately measure visceral adipose tissue (VAT) becomes more imperative as its contribution to disease becomes more clear.

To date, MRI and CT remain the imaging modalities of choice for VAT assessment (10, 11). However, close scrutiny of the literature reveals that there is no clear consensus regarding a standardized protocol for MRI or CT quantification of visceral fat. A myriad of approaches have been utilized in a variety of studies, and some current clinical investigations persist in using outmoded technologies (5, 12). Equally problematic is the fact that the most common methods for assessing VAT by MRI or CT rely on conventional sequences and subsequent computer-generated gray-level histograms with thresholds for water and fat determined from a range of signal intensities (in the case of MRI) or mean attenuation values (in the case of CT) (10, 11, 13). In such studies the lack of a unique or characteristic signal intensity or density for fat creates the potential for water or lean tissue to be incorporated into VAT calculations, rendering quantification unreliable or inconsistent. Although water-saturation MRI sequences have been reported as a technical development in the scientific literature (14, 15), to our knowledge the implementation of spoiled gradient-echo water saturation for the specific quantification of VAT has not been reported. The purpose of the current study was to develop a method that would utilize a water-saturation sequence, to permit the selective display of adipose tissue, in concert with a specially designed computerized segmentation technique to more accurately quantify VAT in the entire abdominal cavity.



Five women of reproductive age were included in the study. The patients were part of a larger study sponsored by the National Institute of Environmental Health Sciences (NIEHS) that was designed to investigate fibroid growth patterns. The NIEHS protocol had prior IRB and HIPPA approval. Additional IRB and HIPPA approval was obtained to acquire images for visceral fat quantification. The procedure and the nature of the study were explained to the study subjects. Informed written consent was obtained. All images were reviewed by experienced body MRI radiologists.


All MRI studies were performed at 1.5 T (Sonata; Siemens Medical Solutions, Malvern, PA, USA). Comparison scans were performed using two sequences: 1) a conventional T1-weighted spoiled gradient-echo breath-hold sequence (non-water-saturation, TR = 285 msec, TE = 4.4 msec, FA = 90°), and 2) a T1-weighted spoiled gradient-echo water-saturation breath-hold sequence (water-saturation, TR = 285 msec, TE = 1.61 msec, FA = 90°). For both sequences images were obtained in two or three interleaved sets, each of which consisted of 16 images. Whole-volume coverage of the abdomen and pelvis, from the top of the liver to the femoral heads, was obtained using 40–45 contiguous transverse slices with 8-mm slice thickness, 2-mm gap, and an image matrix of 154 × 256. The field of view (FOV) varied with the size of the patient, with 400 mm representing the most common transverse measurement. The maximum scan duration (for 48 transverse slices) using water saturation was one minute, and the sequence duration was 20–24 seconds.

Water saturation was achieved with the use of a two-dimensional frequency-selective excitation-spoiling technique whereby the signal of water was selectively excited using a Gaussian pulse with gradient spoiling of the excited signal. One excitation-spoiling pulse was employed per slice loop. This technique was available on the scanner. Prior to the start of this study, one experienced investigator compared water-saturation MRI with fat-excitation MRI. Since water saturation provided better image quality, we chose to employ that technique in the study. Fat excitation is performed with the use of a complex RF pulse design that results in specific frequency-selective excitation of fat without the need to spoil signal from water.

Computer Segmentation of Data

Image Analysis

Data acquired using non-water-saturation and water-saturation MRI sequences were displayed in histogram form. In the first part of the study, computer segmentation of total abdominal adipose tissue was obtained by applying intensity thresholds to both the non-water-saturation and the water-saturation sequences. In this step, no distinction was made between subcutaneous adipose tissue (SAT) and VAT. The SAT was extracted using NLM Insight Toolkit's implementation of the confidence-connected algorithm, which is a region-growing algorithm (16). The volume of VAT was calculated by subtracting the volume of SAT from the total abdominal adipose tissue volume (intensity threshold output).

In the second part of the study, accurate manual tracing of adipose tissue by a trained operator using classic methodology (13, 17) was performed. Data from the manual technique, including volume calculations and time duration of postprocessing, were compared with those obtained by the computer segmentation technique.

In the third part of the study we utilized the computer segmentation technique to perform repeated measurements from the same data set in order to calculate intrastudy variability (expressed as ± percentages). We reevaluated the data from different sessions using different seed points.

Statistical Analysis

Correlation between VAT volumes generated by the computer segmentation technique and manual methods for both sequences was made using Kappa statistics. The standard error of the estimate was determined for intrastudy reliability of the computer segmentation technique.


In the water-saturation sequence, assessment of adipose tissue volume was accurately performed due to the increased signal intensity gap between adipose and non-adipose tissue. VAT volume quantification using the computer segmentation technique for data generated by the water-saturation sequence showed a close correlation with measurements using the manual technique (mean 14.5% error; Table 1, Fig. 1). In the non-water-saturation sequence, the signal difference between adipose and non-adipose tissue was decreased, which led to an ill-defined intensity gap in all cases. Subsequently, in the step involving automated analysis of SAT, the abdominal wall musculature, viscera, and portions of VAT were included in the segmentation. As a result, a disproportionately large amount of tissue designated as SAT was subtracted from total abdominal fat, which resulted in an underestimation of VAT and overestimation of SAT (Fig. 2). This discrepancy most likely accounts for the fact that VAT volumes measured by the water-saturation sequence were consistently greater than those assessed by the non-water-saturation sequence (Table 2). VAT quantification using the non-water-saturation sequence/computer segmentation technique showed poor correlation with measurements using the manual technique (mean 72% error; Table 1). With the non-water-saturation sequence technique, it was difficult to distinguish between fat and solid organs (Fig. 2).

Table 1. Comparison Between Manual Technique and Computer Segmentation Technique: Accuracy Estimation Using Water Saturation (WS) and Non Water Saturation (NWS) Sequences in One Patient
Accuracy estimation
SequenceManual volume measurement (cm3)Computer segmentation technique measurement (cm3)Error
Figure 1.

a: Water-saturation MRI image (L3). b: Same level screenshot computer segmentation technique showing VAT (white) and SAT (green).

Figure 2.

a: Non-water-saturation MRI image (L3). b: Same level screenshot computer segmentation technique showing underestimation of VAT (white) and overestimation of SAT (green).

Table 2. Comparison Between Water-Saturation (WS) and Non-Water-Saturation (NWS) Sequences Using Computer Segmentation Technique for VAT Volume Measurement
Patient numberNumber of slicesWS volume (cm3)NWS volume (cm3)

Histograms generated from the non-water-saturation and water-saturation MRI sequences show that pixel intensities are more clearly separated into low- and high-intensity clusters in the case of the water-saturation sequence (Fig. 3). This clear separation results from the more-accurate data calculation of the water-saturation technique. Furthermore, the separations closely approximate the tracings performed using the manual technique (Fig. 4).

Figure 3.

Histograms of the non-water-saturation MRI sequence (left) and water-saturation MRI (right). The data from the non-water-saturation sequence show a greater number of peaks of signal intensity compared to the water-saturation sequence.

Figure 4.

a: Non-water-saturation MRI. b: Water-saturation MRI using the manual technique.

Correlation between the computer segmentation technique and the manual technique was good (K = 0.8) for the water-saturation computer data and water-saturation manual techniques. The manual tracing accurately delimited VAT (Fig. 4), and compared closely to the water-saturation computer segmentation technique (see Fig. 1). Poor correlation was observed between the non-water-saturation and water-saturation techniques (K = 0.2), and between the non-water-saturation computer segmentation technique and manual techniques (K = 0.2). The mean time required to derive fat volume per section was 10 seconds for the computer technique and 120 seconds for the manual technique. Quantification of VAT in the abdominal pelvic study consisting of 30 sections required five minutes using the computer technique vs. 60 minutes using the manual technique.

Repeated measurements using the computer technique showed a 2.09% mean standard deviation (SD) for the water-saturation sequence and 26.51% for the non-water-saturation sequence (Table 3).

Table 3. Comparison Between Water Saturation (WS) and Non Water Saturation (NWS) Sequences: Variability Estimation in One Patient Using Computer Segmentation Technique
Variability estimation
Number of measurementWS number of voxelsNWS number of voxels
SD (%)2.09%26.61%


There are drawbacks associated with the most commonly used MRI techniques that employ conventional T1-weighted images (non-water-saturation) (10, 13) and computer-assisted protocols for measuring VAT. Fundamental to these techniques is a range of grayscale values. In the case of MRI, emitted signal intensity is reflected in the degree of pixel shading and is expressed in arbitrary units from a minimum value (black on reconstructed images) to a maximum value (white on reconstructed images). The range of arbitrary units that represent adipose tissue is usually determined from inspection of histograms of pixel intensities, which typically show two peaks of gray values belonging to the resonances of fat and water (10, 17). The range of intensities may vary among and within images (17). The signal intensity associated with an “adipose tissue signal” may differ within a scan because of system and object-related imperfections, such as nonuniform excitation and signal detection, magnetic field heterogeneity, vascular flow, effects of respiratory and other types of motion, and partial volume effects (17). Assessment of VAT is particularly affected by partial volume effects because VAT not only borders the muscle and fascia of the abdominal wall, it is also closely aligned with the irregular contours of the intestines (17, 18).

Another theoretical consideration is what is imaged with water saturation. Since water saturation works on the basis of frequency selectivity, the measured entity is fat chemicals—not, strictly speaking, fat tissue. Although we consider it reasonable to assume that they are effectively the same, it should be acknowledged that adipose tissue contains blood and other stromal elements that are not fat chemicals. However, the proportion of these other components is small.

The use of water-saturation MRI offsets many of the above-cited problems by optimizing contrast between tissues and restricting the range of grayscale (pixel) values and signal intensities, thereby producing an image that selectively isolates adipose tissue (14, 15). Water-saturation MRI creates an image that approaches a binary system of signal intensities (specifically, two values: bright and dark). VAT is displayed as bright or high signal intensity that is distinctly separate from all other water-based, lean tissue, which is rendered dark or with low signal intensity, and from air, which is void of signal. Our study shows that water-saturation MRI is superior to non-water-saturation MRI because it creates a unique signal from fat that is distinct from other tissues and air. Other MR techniques, such as fat suppression and CT, do not result in a unique appearance for fat, which intrinsically limits their accuracy. In fat-suppressed MRI, other entities, such as air, have a very low signal, and the signal intensity difference between low-signal tissues (e.g., muscles, liver, and spleen) and very low-signal fat is not great. This circumstance is also seen with CT. Water saturation may be subject to a mild heterogeneity of fat signal, especially when factors that cause great magnetic susceptibility differences (e.g., excessive bowel gas or metallic implants) are present in the imaging field. However, this was not observed in our subject group. Standard T1-weighted imaging is probably not an ideal technique for quantifying fat; however, since it is the most commonly described approach in the literature (13, 17, 18), we believed it would be the most appropriate technique to compare with our approach. In fact, part of the intent of this work was to demonstrate the limitations of using standard T1-weighted imaging. MRI offers an additional advantage over CT for VAT assessment because in MRI the signal of all nonfat entities (including cortical bone and intraluminal gas) is the opposite of signal from fat. It is plausible that an MRI technique such as water saturation, which optimizes tissue contrast, affords less opportunity for compromising valuable anatomic information in both the imaging process and post-imaging computer segmentation. With conventional imaging sequences and computer analysis systems, establishing clear boundaries between VAT of the mesenteries and omentum, and other closely aligned but different tissue types, is technically challenging, leading to results of questionable accuracy (17, 18). Common examples of sequence and post-imaging analysis difficulties include 1) contour merging and boundary consolidation (non-fat tissue may be incorrectly incorporated into VAT calculations), and 2) contour erosion and edge effacement (the actual presence of VAT may go undetected and subsequently be lost to data collection).

Other methods, such as the Dixon technique and fat excitation, can be used to obtain fat-only MR images. We performed a prestudy comparison between fat excitation and water saturation, and decided on the basis of image quality to use water saturation. Fat excitation may fundamentally be a more rapid technique than water saturation, and therefore is worthy of further evaluation in a formal study. Although the original implementations of the Dixon method were time-consuming, recent implementations are much more rapid, and therefore the possibility of using this approach should also be entertained. Conventional non-fat-suppressed T1-weighted imaging was employed in the present study for comparison because it is the most widely described technique in the literature for fat saturation.

Regarding the computer technique employed in this study, the fundamental principle of the technique is the use of an intensity threshold for segmenting tissue and morphological operations to differentiate the type of adipose tissue. Accurate assessment of VAT was predicated on the use of the water-saturation technique because it increases the signal difference between adipose and non-adipose or lean tissue. This approach differs from previous methods in that it has the potential to perform VAT extraction without human supervision. Its simplicity makes it time-efficient in comparison to sophisticated segmentation techniques that require more computation time and more careful parameterization. Furthermore, this technique showed excellent correlation with the standard manual technique and exhibited low intrastudy variability.

Although in recent years CT and MRI techniques to assess intraabdominal adipose tissue have become available, the heterogeneity of these approaches has precluded the collection of generalizable data. Lack of standardization is related to many factors, including 1) variations in methods of sectioning, such as single transaxial slice imaging at a predetermined site (L4- L5 or L2- L3), multislice transaxial imaging of the entire abdomen, and whole-body transaxial imaging (12); 2) variations in slice thickness and intersegmental gaps; and 3) different methods of quantification (i.e., VAT area (cm2) vs. VAT volume (cm3)). Consistent and reliable data from imaging studies are necessary to generate optimal correlations between clinical parameters and VAT quantifications.

In the past, most investigations used single-slice CT or MRI to measure the cross-sectional area of VAT (19). Single-slice techniques appear to be satisfactory for obtaining an approximate estimate of internal fat or assessing changes in intraabdominal fat content in interventional, longitudinal studies in which patients can serve as their own individual controls (12). However, for studies that necessitate intersubject comparison or determination of the relationship between VAT and biologic and genetic markers, the most accurate measurement of individual VAT content requires whole-body adipose tissue topography assessment by MRI (20), or whole-abdomen multislice MRI (12). The latter approach was recently used by Tintera et al (21) in a water-suppression MR technique to quantify intraabdominal fat and correlate it with laboratory parameters during a weight loss program. The results of that study showed an average relative loss of visceral fat (20.3%) that was greater than both average weight loss (8.2%) and decrease in subcutaneous fat (13.4 %) after controlled weight reduction. It is important to note that Tintera et al (21) employed a spin-echo technique, which involves substantially more time-consuming data acquisition compared to the expeditious acquisition of data afforded by spoiled gradient-echo sequences, as used in the present study. In the spin-echo technique, factors involved in the time-intensive nature of data acquisition include the necessity for individual breath-holds for each slice. This is in sharp contrast to the more efficient data acquisition of the spoiled gradient-echo technique, in which 16 slices are acquired in a single breath-hold. Other considerations when comparing the two techniques are 1) the potential for slice misregistration due to variations in breath-holding, and 2) the time interval between breath-holds, which further increases the duration of the study.

Perhaps the most compelling advantage afforded by MRI over CT in the evaluation of VAT is the inherent safety of the modality. Because of health risks associated with exposure to ionizing radiation, CT is not an optimal imaging modality for VAT assessment, and is especially unsuitable for studies that require repeated measurements on the same patient (22–27).

Most prior clinical investigations involving assessment of intraabdominal fat used the ratio of intraabdominal visceral fat (V) to the subcutaneous fat (S) area, or V/S, as a relative index of intraabdominal fat accumulation (4). In the present study we chose to focus exclusively on the VAT depot based on the well-established, strong correlation between intraabdominal adipose tissue accumulation, or visceral obesity, and IR. VAT possesses distinctive metabolic properties compared to other adipose tissue. The rate of lipolytic activity in visceral fat cells is higher than that in subcutaneous fat cells, which favors mobilization of free fatty acids to the portal vein as compared to peripheral veins (28). An excess of visceral fat has deleterious effects on liver metabolism that can lead to fatty liver, IR, and elevated risks of cardiovascular disease and type 2 diabetes (29). Research studies have reinforced the finding that VAT alone is a strong correlate of IR, independently of nonabdominal and abdominal SAT (30). Recent evidence points to visceral fat as the most important physical characteristic associated with metabolic syndrome (31).

Research clearly demonstrates that individuals with visceral abdominal obesity are more prone to develop metabolic and cardiovascular disorders (4–9). Weight loss, by any means, results in loss of VAT (19). Decreases in VAT contribute to improvements in certain components of metabolic syndrome, and thus potentially decrease the risk of cardiovascular disease (32). The ability of therapeutic interventions such as diet, exercise, or pharmacologic agents to selectively target VAT will become increasingly important in the years to come. This has significant implications for MRI, since accurate and reliable methods for quantifying VAT will be critical for generating consistent and generalizable data. Further, more research is needed to determine whether measurement of visceral fat by imaging studies could be used by physicians as a routine diagnostic tool to more accurately predict the cardiovascular risk posed by excess weight (33).

A limitation of this study is the small number of studies performed. Future validation of the technique will require a larger cohort of patients.

In summary, our preliminary results from a limited number of patients suggest that the application of a water-saturation MRI sequence to specifically delineate adipose tissue, combined with a specially designed computer analysis tool, shows potential for more accurately measuring intraabdominal adipose tissue. This study represents the first step in the maturation of an MRI modality in becoming the gold standard for rapid, reliable, and reproducible VAT quantification, which can be performed both accurately and safely.