Assessment of abdominal adipose tissue and organ fat content by magnetic resonance imaging

Authors


HH Hu, University of Southern California, 3740 McClintock Avenue, Electrical Engineering Building (EEB) 408, Los Angeles, CA 90089-2564, USA. E-mail: houchunh@usc.edu

Summary

As the prevalence of obesity continues to rise, rapid and accurate tools for assessing abdominal body and organ fat quantity and distribution are critically needed to assist researchers investigating therapeutic and preventive measures against obesity and its comorbidities. Magnetic resonance imaging (MRI) is the most promising modality to address such need. It is non-invasive, utilizes no ionizing radiation, provides unmatched 3-D visualization, is repeatable, and is applicable to subject cohorts of all ages. This article is aimed to provide the reader with an overview of current and state-of-the-art techniques in MRI and associated image analysis methods for fat quantification. The principles underlying traditional approaches such as T1-weighted imaging and magnetic resonance spectroscopy as well as more modern chemical-shift imaging techniques are discussed and compared. The benefits of contiguous 3-D acquisitions over 2-D multislice approaches are highlighted. Typical post-processing procedures for extracting adipose tissue depot volumes and percent organ fat content from abdominal MRI data sets are explained. Furthermore, the advantages and disadvantages of each MRI approach with respect to imaging parameters, spatial resolution, subject motion, scan time and appropriate fat quantitative endpoints are also provided. Practical considerations in implementing these methods are also presented.

Introduction

The prevalence of obesity has risen steadily (1). Literature evidence has attributed obesity-related health risks to the accumulation of subcutaneous and visceral adipose tissue (SAT, VAT) and ectopic organ fat (2–4). Non-invasive assessment of body fat has become an important component in obesity research and fat measures are useful as biomarkers to stratify risks and evaluate the efficacy of therapies.

Many methods are available for body fat assessment (5,6). Anthropometry, hydrodensitometry, air displacement plethysmography (ADP), bioelectric impedance (BIA), and dual energy X-ray absorptiometry (DEXA) have been widely used. The first four are indirect techniques because they measure body density or resistance, which are then converted into percent body fat using generalized equations (7). Hydrodensitometry and ADP only estimates total body fat. The BIA and DEXA techniques are limited to total and regional body fat measures. Indirect methods are not able to further differentiate between SAT and VAT or identify ectopic fat.

Recent years has seen an increased utilization of computed tomography (CT) (8–10), quantitative magnetic resonance (QMR) (11,12) and particularly magnetic resonance imaging (MRI) for body fat assessment (13–24). Like DEXA, these are direct techniques because they identify fat based on the tissue's unique properties. QMR does not yield images and is limited to measuring total body fat mass. In contrast, CT and MRI can differentiate SAT and VAT with multidimensional images. However, only MRI, with its many sensitive and flexible contrast mechanisms, can quantify ectopic fat. MRI involves no ionizing radiation, which allows for indefinite repeatability in longitudinal studies and in children. It is emerging as a comprehensive tool for fat quantification (25).

This article reviews state-of-the-art MRI methods and image analysis tools for body and organ fat quantification. A summary of MRI of fundamental concepts that underpin T1-weighted imaging, magnetic resonance spectroscopy (MRS), and chemical-shift imaging (CSI) is given. Relevant examples are shown using data acquired from a 3 Tesla General Electric scanner (GE Healthcare, Signa HDx 14M5), unless otherwise stated. General procedures for extracting adipose tissue volumes and organ fat measures from abdominal data are discussed and illustrated using commercial software (SliceOmatic, Tomovision, Inc.). The advantages and disadvantages of MRI approaches with regards to imaging parameters, subject motion, scan time and quantitative endpoints are highlighted, along with practical considerations in implementation. The article is presented in the context of abdominal imaging, but the content can be translated to other anatomies.

Magnetic resonance imaging methodology

Magnetic resonance imaging signal formation

The MRI signals arise from the magnetization of nuclei when placed in a BO magnetic field. Hydrogens, or spins, of water and fat are the most commonly measured nuclei. Today's clinical MRI scanners utilize a BO of 1.5 or 3 Tesla to align spins and generate magnetization. The spins rotate at a characteristic resonant frequency that is proportional to BO. Radiofrequency (RF) pulses tuned to this frequency are then used to repetitively excite the magnetization by a specified flip angle. Signals are acquired from a RF pulse sequence by receivers as the excited magnetization decays, or relax, between successive pulses. The relaxation rates are known as T1 and T2 times (26).

T1 weighting

The MRI signal intensities are influenced by many factors including spin density, tissue-specific T1 and T2 rates, the flip angle, and the repetition time of the pulse sequence. The T1 of fat is one of the shortest in vivo, which indicates a very rapid recovery of the magnetization between successive RF excitations. By using a T1-weighted sequence, strong tissue contrast can be achieved between short-T1 fat and muscles and organs with longer-T1 values (Fig. 1). Thus bright fat can be easily identified and delineated by simple signal thresholding from darker structures. T1-weighted sequences are very common in clinical MRI. They are available as standard software on all commercial MRI scanners (GE, Philips, Siemens, Hitachi, and Toshiba) and can be easily implemented. Tissue T1 values typically increase with BO field strength, such that slight adjustments in pulse sequence parameters are needed to maintain comparable tissue contrast between 1.5 and 3 Tesla.

Figure 1.

Examples of T1-weighted axial magnetic resonance images acquired in the (a) abdomen and (b) thigh, demonstrating the typical high signal intensities of fatty tissues (arrows) in contrast to other darker muscles and organs (L, liver; P, pancreas; K, kidneys; M, muscle). Data were acquired on a GE 1.5 Tesla scanner (Signa HD, 12M5).

Magnetic resonance spectroscopy

Single-voxel MRS has been the standard for ectopic fat quantification (27–32). The MRS technique does not provide anatomical information. It instead yields a precise spectrum of chemical composition within one interrogated voxel. The MRS method relies on chemical shift, which refers to differences in the resonant frequencies of fat and water spins. Fat spins are characterized by a spectral peak that is offset from water spins. Because of differences in their chemical surroundings, lipid spins have a lower resonance. At 1.5 and 3 Teslas, the water-fat chemical shift is approximately 210 and 420 Hz, respectively (14,33–35). The MRS spectrum provides an intuitive visualization of the presence and relative quantity of chemical species. It also benefits greatly from increasing BO fields because of larger chemical-shift separations between peaks. The MRS packages should be available on most commercial MRI scanners as standard or optional software but usually requires some expertise to implement and analyze.

Frequency-selective magnetic resonance imaging

Because MRI involves RF pulses tuned to specific frequencies, techniques that exploits the resonant frequency difference between fat and water to quantify SAT and VAT has been developed. One approach is to selectively excite fat spins by tuning the RF pulses to their resonant frequency while simultaneously suppress the water signals (36–38). The appearance of frequency-selective images resembles T1-weighted results and the sharp contrast between fat and musculature readily facilitates thresholding procedures. Frequency-selective techniques can be implemented on all MRI scanners although the capability may not be included as standard software. The technique is sensitive to spatially varying BO inhomogeneities, which worsens with increasing magnetic field. Despite greater chemical-shift separation between fat and water at higher field strengths which allows for easier frequency selectivity, BO inhomogeneity can significantly degrade its performance.

Chemical-shift magnetic resonance imaging

The CSI method integrates fat-water spectral detection with imaging formation (39). The term CSI represents a family of approaches that have been developed over the past 25 years. Dixon was the first to demonstrate that by controlling the time when data was acquired after RF excitation (e.g. echo time), the net detected MRI signal can comprise either of fat and water in-phase (W + F, aligned) or out-of-phase (W–F, anti-aligned), as illustrated in Fig. 2(40). Using this 2-point approach, separated fat and water images could be obtained by algebraic manipulation (34). Intuitively, one realizes that for a voxel containing only fat or water, its net signal will be the same on in-phase and out-of-phase acquisitions as one component's signal will be zero. In contrast, a voxel containing fat and water will have different signals on the two acquisitions. With higher BO field strengths, the echo time separation between in-phase and out-of-phase signals decreases from every 2.4 ms at 1.5 Tesla to every 1.2 ms at 3 Tesla.

Figure 2.

Example of 2-point chemical-shift imaging technique. In-phase (IP) and out-of-phase (OP) images from two subjects are shown. In the OP images, a dark line is present at all fat and lean tissue interfaces. This is caused by signal cancellation between fat and water species within the voxels. All IP and OP images are displayed on the same greyscale. In (a), signal intensities within the liver between the two images are similar, indicating very little presence of hepatic fat. However, in (b), the liver signal intensity is markedly lower in the OP image (dashed region), indicative of fat presence. This is visualized in the zoomed difference images. The liver in (a) exhibits residual signal whereas that in (b) shows appreciable signal. As corroborated by magnetic resonance spectroscopy, the hepatic fat fractions were 3.4% and 21.2%, respectively.

The CSI set of techniques has evolved over the past decade (34,41,42). More robust protocols have been developed to address BO inhomogeneity, which results from manufacturing imperfections, perturbations of the magnetic field by placement of a human body, and at air-tissue-bowel interfaces. Modern CSI approaches (43) also address fat-water signal ambiguity, which refers to the inability of previous approach to detect whether fat or water is the predominant specie present in a voxel. A voxel contain x% fat and (100 − x)% water can be indiscernible from a voxel containing x% water and (100 − x)% fat (e.g. there is ambiguity about the 50% fat fraction apex). Many literature reports have used CSI to measure organ fat fraction (29,43–48). While not difficult to implement, they often require technician involvement and manufacturer support in establishing appropriate imaging parameters. The CSI pulse sequences and reconstruction software (GE – Lava Flex, IDEAL, Philips – mDixon, Siemens – Dixon, Hitachi – FatSep) may only be available on certain new commercial MRI models as standard clinical or optional research-dedicated software.

Recent advances in CSI have led to the development of comprehensive methods such as IDEAL (Iterative Decomposition with Echo Asymmetry and Least squares) (49). Along with several variants (50,51), IDEAL is a generalization of the multiecho Dixon model and it has been demonstrated at both 1.5 and 3 Tesla. At both field strengths, IDEAL is robust to BO inhomogeneity, addresses fat-water signal ambiguity such that fat fractions can be measured uniquely across the 0–100% range (52,53), accounts for relaxation (54,55) and models the multiple spectral peaks of fat (56). A single IDEAL acquisition yields fat-only and water-only images and fat fraction maps (F/[F + W]) that are optimal in signal-to-noise ratio (57) (Fig. 3), as well as in-phase and out-of-phase series. It has been validated against MRS and other CSI methods (29,48,58). IDEAL has also been used to study brown and white adipose tissues (59), spine (60) and knee cartilage (61,62). To the author's knowledge, IDEAL is a patented and proprietary algorithm that is commercially available only from GE Healthcare on their Discovery 1.5 and 3 Tesla platforms and as a research software on the Signa HDx platforms.

Figure 3.

Examples of 3 Tesla IDEAL. Reconstructed water-only, fat-only, and fat fraction images are illustrated for the thigh and upper abdomen. In the colour fat fractions, the scale represents 0–100% fat content. In the thighs, note that subcutaneous and intramuscular adipose tissues and bone marrow are denoted by high (>90% red) fat fractions. In the liver example, the colour representation is indicative of very high 40% hepatic fat content. Coronal reformat illustrates the entire liver. The arrow in the water image denotes ripple-like artefacts from respiratory motion.

Summary

The MRI and MRS approaches provide a vast array of sensitive methods for quantifying fat (Table 1). The interested reader is referred to three extensive review articles in recent literature for additional detailed technical explanations (34,41,42).

Table 1.  Summary of magnetic resonance imaging and spectroscopy methods for fat quantification
 Basis of fat-water differentiationTypical dimensionalityAdvantagesDisadvantagesQuantitative endpoints
  1. MRI, magnetic resonance imaging; MRS, magnetic resonance spectroscopy; SAT, subcutaneous adipose tissue; VAT, visceral adipose tissue.

T1-weighted MRIT1 relaxation2-D and 3-DStrong fat-water tissue contrast, standard on all scanners, easy to implement, rapid scansT1 contrast varies with BO field strength, requires protocol optimizationSAT and VAT volumes
Frequency-Selective MRIChemical shift2-D and 3-DStrong fat-water tissue contrast, option on most scanners, rapid scansPerformance varies with BO field strength and magnet inhomogeneitySAT and VAT volumes
MRSChemical shiftSingle voxelHigh spectral resolution of chemical profile, very accurate measure of fat to water fractionsLimited coverage, requires expertise in protocol prescription, data collection and spectral analysisEctopic fat fraction
Chemical-shift MRIChemical shift2-D and 3-DComplete fat-water tissue separation with accuracy comparable to MRSNot available on all scanners, requires special image reconstruction algorithms and protocol optimization for different Bo field strengthsSAT and VAT volumes, ectopic fat fraction

Image analysis

The workflow for extracting fat volumes and organ fat fractions from MRI data remains a daunting and costly task. It requires substantial data post-processing and analysis by an experienced operator. The procedure involves the transfer of images to an offline workstation, followed by the use of dedicated software that requires manual intervention (63). Free packages such as ImageJ and Osirix and commercial programs such as SliceOmatic (Tomovision, Inc.), Analyze (AnalyzeDirect, Inc.) and Matlab (The Mathworks, Inc.) are commonly used for segmentation. The analysis of MRS data requires additional software (64). In our experience, post-processing of a 3-D abdominal MRI volume takes 45–60 min.

Subcutaneous and visceral adipose tissue

Because adipose tissue is composed primarily of lipids, their identification and quantification in T1-weighted, frequency-selective, and chemical-shift images is straightforward. The MRS technique can not be used because it does not provide anatomical data. For T1 and frequency-selective approaches, a binary threshold is usually applied to the signal intensities of each voxel. Because fat appears brighter, those with intensities greater than a set threshold are labelled as fat while those with lower intensities are excluded. Because of intensity variations that may arise across the image, the local threshold may need to be adjusted accordingly (Fig. 4). For CSI data, the analysis can utilize the fat-only images and apply a similar threshold procedure. Alternatively, the analysis can exploit the fat fraction map where the range is consistently fixed from 0–100%. Only voxels with fat fractions greater than a set value (e.g. 80%) are then labelled as fat.

Figure 4.

Examples of SliceOmatic segmentation to delineate the subcutaneous (red) and visceral (green) adipose tissue depots. Note the evident difference in visceral adipose tissue quantity between the two subjects. T1-weighted in-phase (left column) and out-of-phase (middle column) images are shown. Labels (right column) were generated from in-phase images for volume accuracy, using the out-of-phase image for fat-muscle boundary guidance. Arrows highlight bright intensity regions caused by close proximity of the anatomy to receiver arrays. Note that bowels, intramuscular fat, blood vessels, bone and spine-vertebrae structures are excluded from visceral fat segmentation.

Regardless of the MRI technique, the operator must also delineate the intra-abdominal boundary that separates SAT and VAT and the perimeter of the abdomen to exclude background. Additional exclusions of bowel contents, intramuscular fat (65), spine and vertebrae, blood vessels and other non-relevant structures are also required since some of these can have signal intensities that mimic fat. An area or volume measure is then computed for 2-D multislice and contiguous 3-D data based on the identified fat voxels, respectively. With the exception of CSI fat fraction data that is always normalized between 0% and 100%, the operator must determine suitable thresholds on a subject-to-subject basis. This is necessary since signal intensities vary across anatomy, subjects, examinations and equipment and is usually not consistent. In SliceOmatic and Analyze, the operator has control over a sliding threshold and an interface that provides visual feedback on the performance of the chosen threshold. Figures 4 and 5 illustrate multiple examples.

Figure 5.

Examples of SliceOmatic image segmentation from two subjects using IDEAL-reconstructed data, one with a fatty liver (top row) and one with a non-fatty liver (bottom row). Water-only, fat fraction, and overlay of segmentation labels (red: subcutaneous adipose tissue, green: visceral adipose tissue, blue: liver) are shown in the left, middle, and right columns, respectively. For illustration purpose, the fuzzy noisy background in the fat fraction images has not been removed. In the fatty liver subject, note that hepatic vessels and the gall bladder (blue arrows) are well delineated in both water and fat fraction images, whereas in the non-fatty liver subject, such structures are only observed in the water image. Note that corresponding segmentation labels of the liver exclude these non-liver-tissue structures. Similarly, visceral adipose tissue labels do not include erroneous signals from empty bowels (green arrows).

Ectopic fat

Ectopic fat quantification requires approaches that can separate signals from lipid and non-lipid components within each imaging voxel. Therefore, the aforementioned binary threshold approach of ‘all fat’ or ‘no fat’ voxel classification is inadequate and only CSI methods are appropriate. Rather than measuring volume, a fractional estimate of fat and water is computed. For MRS, the areas under the fat and water spectral peaks represent the relative amounts of each species present. Software such as jMRUI allows the operator to determine each peak's area efficiently. The MRS fat fraction is then defined as the ratio of area under the fat peaks to the combined area of the fat and water peaks (48).

The CSI methods can provide a fat fraction map across the imaging volume. The map is computed from the ratio of the reconstructed fat-only images to the sum of the fat and water images. To determine the fat fraction at a particular location, the operator needs to draw a region-of-interest (ROI). The statistical distribution of fat fractions within the ROI (e.g. mean, standard deviation) is then computed. To further quantify organ fat fraction, the operator needs to manually segment the entire organ. Once segmented, an overall average fat fraction across the organ can be computed. Unlike single-voxel MRS, a volumetric fat fraction map affords the operator immense flexibility. One can determine the average fat fraction across any arbitrary ROI, including different lobes of the liver, the head and the tail of the pancreas, and across muscles.

For accurate fat fractions, operator expertise is required for drawing ROIs and segmenting organs. In some cases, the fat fraction signal contrast surrounding abdominal organs may be poor, limiting an operator's ability to directly segment structures from the fat fraction data series and visualize organ–muscle boundaries. This is particularly true for liver, pancreas, and kidney with low-fat content that appear visually similar to surrounding muscles. In these instances, reference greyscale image series, including water-only, fat-only, in-phase and out-of-phase data, are needed to provide anatomical landmarks. The operator can then segment from these greyscale images and transfer the ROIs to the corresponding fat fraction data (Figs 5 and 6). In our experience, manual analysis of organ fat accounts more than 70% of the post-processing time. The pancreas and the kidneys are also in close proximity to VAT depots. An inaccurately drawn ROI meant to enclose the organs may contain erroneous VAT voxels. In such cases, the high VAT fat fractions will significantly impact the apparent measured organ fat fraction.

Figure 6.

Manual image segmentation using SliceOmatic and the full spectrum of IDEAL data to further delineate abdominal organs (red: subcutaneous adipose tissue, green: visceral adipose tissue, blue: liver, yellow: pancreas, purple: kidneys). Left to right: water-only, fat-only, in-phase, out-of-phase, fat fraction and segmented labels. Labels were generated from the first four columns of source greyscale images.

Summary

Post-processing by a trained operator is needed for computing quantitative fat endpoints. For SAT and VAT volumes, strong tissue contrast can facilitate signal-intensity-based thresholding. With the aid of software, this procedure can be efficiently performed. The measurement of ectopic fat requires MRS and CSI. Manual segmentation of organs demands familiarity with software and anatomy, as well as significant operator input.

Discussion

An investigator has several choices in MRI for body fat assessment. In addition to selecting the appropriate technique, one must also consider spatial resolution, scan time, and the need for breath-holds to minimize subject motion (Fig. 7).

Figure 7.

Practical imaging parameters and considerations in magnetic resonance imaging. In designing an appropriate protocol, one must consider the desired quantitative endpoints and leverage spatial resolution, scan time, patient safety and the need for breath-holding.

Spatial resolution describes the dimensions of the imaging voxel. For MRS, 1–8 cm3 is commonly used. MRS requires a large voxel and several signal averages to yield a clean spectrum. Multiple scans are also needed to account for T1 and T2 relaxation (66). The MRS voxel prescription process requires operator expertise such that placement of the voxel does not contain any undesired tissues. Motion from the subject or shifting of organs because of respiration or peristalsis after voxel placement can lead to a spatial misalignments (48). Respiration should be monitored with a bellows transducer for data synchronization. Alternatively, multiple and consistent breath-holds can be used (31). In practice, these approaches may not be possible in subjects with a large abdominal circumference or who have difficulties with breath-holds.

Spatial resolutions for imaging approaches are on the order of millimetres. While small voxels are desirable for visualizing anatomical detail, it leads to increased scan times and lowers image signal-to-noise ratio. Long scan times also increase the risk of motion and blurring artefacts. In CSI, blurring artefacts from respiration can further cause unrealistic fat fraction distributions within organs and at organ-fat interfaces. High spatial resolution scans also necessitate breath-holding, which may not be well tolerated by children, uncooperative or obese subjects. Spatial resolution is determined by the field-of-view and the acquisition matrix, the number of slices, the slice thickness, and the inter-slice gap, if any. These parameters are freely chosen by the operator and are dependent on body habitus.

The operator must select between 2-D multi and slice or 3-D volumetric pulse sequences. With 2-D techniques, data are acquired on a slice-by-slice basis such that data obtained for one slice is independent of any other slice. Subject motion and artefacts are generally not an issue. The time needed to acquire each slice is usually very short, requiring subseconds to a few seconds. The technique also allows the operator to specify a gap between non-contiguous adjacent slices. While this appears attractive in terms of scan efficiency, large interslice gaps and thick (∼10 mm) slices can lead to partial volume errors and impact quantitative accuracy.

In contrast, 3-D approaches acquire data that represent the entire volume. Volumetric imaging implies contiguous coverage and no gaps. Given similar imaging parameters and coverage, 3-D protocols often contain more slices, have longer scan times, and are more susceptible to motion than 2-D approaches. However, because signals arise from a volume rather than a slice, 3-D methods exhibit greater signal-to-noise ratios. Breath-holding or respiratory gating is often required for abdominal 3-D sequences. Three-dimensional approaches offers intrinsic advantages over 2-D methods, such as exclusive parallel imaging acceleration, greater acceleration rates, and more robust parallel imaging performance to reduce scan time (67,68). Other 3-D benefits include acquiring thinner slices and accommodating preparatory RF pulses that further enhances T1 tissue contrast. These advantages largely arise from hardware and RF tissue heating and safety limitations that are beyond the scope of this article.

Lastly, for a given set of resolution and parameters typically used in abdominal imaging, scan times in T1-weighted, frequency-selective, MRS and CSI approaches should be comparable between 1.5 and 3 Tesla protocols. In some instances, safety and RF tissue heating limits may lengthen 3 Tesla protocol durations over their 1.5 Tesla counterparts. At higher BO field strengths, greater chemical-shift differences can also cause CSI sequences to increase in scan time because of shorter in-phase and out-of-phase echo periodicities.

In the literature, 2–4 mm in-plane resolution and a slice thickness between 5 and 10 mm have been reported for SAT and VAT quantification. For 2-D multislice approaches interslice gaps as wide as 10 mm have been reported. Since the determination of volumes rely on thresholding and an ‘all fat’ or ‘no fat’ labelling each voxel, any incorrect labelling can lead to quantification errors when using large voxels. Large voxels also generate partial volume effects where many can encompass both fat and lean tissues. These voxels exhibit ambiguous signal intensities that can cause threshold errors. Similarly, low spatial resolution and thick slices containing both organ and adjacent fat can exhibit unrealistic CSI fat fractions. There has been a transition in recent studies to adopt contiguous 2-D multislice approaches with thin slices (22) and 3-D protocols (24,69,70). While ectopic fat quantification with 3-D IDEAL (Fig. 8) has been described, its use in large cohorts has yet to be reported because of the method's limited availability.

Figure 8.

Abdominal fat fraction volume from IDEAL, illustrating the contiguous 3-D nature of the data. The volume consists of more than 70 native axial slices and was obtained with five 15 s breath-holds (R, right; L, left; A, anterior; P, posterior; I, inferior; S, superior). Rendering created with 3DSlicer software.

Lastly, modern MRI platforms are equipped with receiver arrays designed specifically for abdominal parallel imaging (68). The array is strapped to the subject to provide high signal-to-noise ratios. However, they also create signal intensity variations across the imaging volume (Fig. 4). If possible, these variations should be corrected with software from MRI vendors prior to image analysis. One attractive feature of fat fraction maps is that they are generated as a ratio of fat and water images. While the individual source images may exhibit signal variations, their ratio removes the bias. The use of arrays may be limited in obese subjects as they are unable to fit within the magnet bore with the receivers strapped.

Summary and future directions

The design of appropriate MRI protocols for fat assessment requires considerations of the quantitative endpoints and tradeoffs in the imaging parameters. Other practical issues include body sizes of the cohort and the cost and throughput efficiency of the study. The equipment cost associated with MRI can range from $300–600 per hour in a typical clinical radiology department. Regardless of the study, overhead time is always required for setup and for giving instructions to the subject. T1-weighted and frequency-selective approaches are very fast and typically require only a few seconds per slice for data acquisition. Whole-abdomen coverage with these sequences takes only a few minutes of actual scan time and is generally considered as a first option for SAT and VAT quantification. In our experience, utilizing only these sequences, an exam can be comfortably completed in 10 min. If ectopic fat fractions are needed, either a MRS or CSI protocol is then added. The MRS sequence necessitates significant user interaction, preparation work, and operator time. In our experience, MRS per single voxel takes approximately 5 min. The further addition of a 3-D whole-abdomen CSI protocol such as IDEAL will require 5–7 more breath-holds, bringing the total examination time from setup to patient removal to 30–45 min.

Despite its rapid growth, MRI remains an untapped resource in obesity research and has not yet reached its full potential. Many opportunities remain for investigation. Several groups have recognized the time-consuming and labour-intensive work involved with manual image segmentation. This can be daunting in longitudinal studies, as the process must be performed for every scan at multiple time points across the same subject. Efficient semi-automated and automated algorithms have been proposed, but are usually limited to in-house usage (19,38,71–75). Techniques that can automatically segment adipose tissue and organs from a subject's MRI data at subsequent time points while utilizing a priori information from the same subject's baseline segmentations are being explored. The capability of achieving rapid intrasubject 3-D registration, segmentation, and quantification will provide investigators with detailed person-specific information reflecting the temporal change in fat distribution and volumes in response to intervention. Such capability will also exploit the richness of 3-D MRI that is unavailable in gapped 2-D multislice data. Quality control also will be essential in assuring consistent accuracy. Fat measurements should be routinely validated against phantoms, between multiple blinded observers, and between manual and automated segmentations.

In conclusion, MRI is the most powerful and comprehensive imaging tool for fat quantification. With CSI methods, recent studies are beginning to demonstrate their usefulness and explore new fat biomarkers in obesity. This article has provided introductory materials for investigators to survey and integrate MRI into their body composition studies.

Conflict of Interest Statement

No conflict of interest was declared.

Acknowledgements

The authors gratefully thank Huanzhou Yu and Ann Shimakawa from GE Healthcare for providing MRI technical support, and acknowledge grant support from the National Institutes of Health (Grant numbers: U54CA116848, R21DK081173, K25DK087931). The authors also thank Rosa Rangel and Sherryl Esplana from the University of Southern California for research assistance, as well as Mark Punyanitya from Columbia University for guidance with image segmentation.

Ancillary