Precise, noninvasive analysis and quantification of in vivo body composition is essential for research involving longitudinal, small-animal disease models. We investigated the feasibility and precision of a rapid, flat-panel μCT scanner to report whole body adipose tissue volume (ATV), lean tissue volume (LTV), skeletal tissue volume (STV), and bone mineral content (BMC) in 25 postmortem female and 52 live male Sprague-Dawley rats. μCT images, acquired in three 90-mm segments and reconstructed with 308 μm of isotropic voxel spacing, formed contiguous image volumes of each entire rat specimen. Three signal-intensity thresholds (determined to be −186, 5, and 155 HU) were used to classify each voxel as adipose, lean, or skeletal tissue, respectively. Tissue masses from the volume fractions of ATV, LTV, and STV were calculated from assumed tissue densities of 0.95, 1.05, and 1.92 g/cm−3, respectively. A CT-derived total mass was calculated for each rat and compared with the gravimetrically measured mass, which differed on average for the postmortem female and the live male group by 2.5 and 1.1%, respectively. To evaluate the accuracy of the CT-derived body composition technique, following the live male study excised muscle tissue in the lower right leg of all rats in group B were compared with the image-derived LT measurement of the same regional compartment and found to differ on average by 2.2%. Through repeated CT measurements of postmortem specimens, the whole body ATV, LTV, STV, and BMC measurement analysis gave a precision value of ±0.6, 1.9, 1.7, and 0.5% of the average value, respectively.
- microcomputed tomography
- body composition
- bone mineral content
body composition analysis is important to both human and small-animal disease investigations and in the direct observation of disease treatment. These can include investigations of obesity, diabetes, osteoarthritis, and osteoporosis and may include a broader range of disease models where more than one tissue type may be affected in unexpected ways. Therefore, it is important to develop methods to accurately measure the in vivo volume fraction of the three dominant contributors to total body composition in mammals: adipose tissue, lean tissue (muscle), and skeletal tissue (ST) (inorganic and organic components). Furthermore, techniques used to measure these quantities must be sufficiently precise so that small changes in the fractional proportion of the tissue volumes can be observed and monitored in longitudinal small-animal studies. Many techniques have previously been developed to measure body composition in humans and in small animals, but most are either indirect measurements or invasive and destructive methods (12, 14, 21, 34).
Indirect techniques like underwater weighing (body buoyancy) and bioelectrical impedance, although easily repeatable, ultimately require cross-validation with direct methods and are, therefore, less sensitive than direct methods (3, 12, 34). They are also limited due to the fact that they are all global methods and do not provide localized body composition information in an anatomic region of interest; for example, they could not quantitatively measure longitudinal changes in subcutaneous or visceral adipose tissue. In particular, spatial characterization of adiposity is important because it is known that there are differential physiological risks associated with anatomic region (9, 13, 18). Chemical analysis is the gold standard technique and the most direct and accurate method for obtaining body composition, but this is highly destructive, precluding its use clinically and in longitudinal studies of small animals.
Noninvasive in vivo imaging techniques such as magnetic resonance imaging (MRI), dual-energy X-ray absorptiometry (DEXA), and computed tomography (CT) have been shown to accurately quantify adipose (AT) and lean tissue (LT) compared with the gold standard (22, 27, 28) and are now commonly used in clinical body composition analysis. Quantitative nuclear magnetic resonance has also been found to be useful in characterizing in vivo the total body composition of small animals (32). However, MR-based techniques, which use hydrogen to probe small-animal specimens in vivo, suffer in water-depleted regions such as bone, making routine bone segmentation from MRI images difficult without using highly specialized, large-angle, fast-spin echo sequences (35). DEXA is the most widely used bone density measurement technique but provides only limited spatial localization because it classifies tissue components on the basis of a two-dimensional projection of the anatomy of interest. CT, which forms three-dimensional (3D) images of tissues, has the ability to measure highly localized regions of AT or LT and has long been established as a clinical tool for calculating bone mineral density (referred to as quantitative CT) (19, 28, 31). Microcomputed tomography (μCT) has also now been used to quantify regional tissue volumes of subcutaneous and visceral adipose in mice (15, 19).
Recent developments in μCT scanner technology will make body composition analysis of small animals using μCT more practical; for example, a recently developed in vivo cone beam μCT scanner (eXplore Locus Ultra; GE Healthcare, London, ON, Canada) can acquire high-resolution, 3D, volumetric images of live small animals, as large as a rabbits, in seconds. The nominal resolution offered by this μCT scanner (154 μm isotropic voxel spacing), in contrast to clinical scanners (nominal 625 μm), provides sufficiently high spatial resolution to minimize partial-volume averaging effects when regional tissue volumes are calculated on the basis of thresholding techniques (7, 29). μCT imaging also provides quantitative data that can be rapidly acquired, dramatically reducing the time and cost for whole body animal composition analysis.
We present μCT images of both in vivo and ex vivo whole body rats postprocessed to derive the total volume of AT, LT, and ST (inorganic and organic bone components) on the basis of CT grayscale intensity. Two small-animal research studies investigating a prevalent disease and its treatment, one of an osteoporosis model (group A) and the other of a diabetic model (group B), were selected to demonstrate a range of body composition values that have previously been known to exhibit perceptible changes to total body mass (10). Voxels characterized as skeletal tissue were processed to calculate the whole body bone mineral content (BMC) for each rat. To calculate the precision of the cone beam μCT imaging system to report BMC and tissue volume measurements, a subset of the rats was scanned multiple times postmortem. To evaluate the accuracy of the tissue mass measurements, excised muscle compartments in the lower right leg were weighed and compared with the corresponding reconstructed image region CT-derived mass. Furthermore, by μCT imaging of the entire animal, we are able to calculate a CT-derived estimate of the total weight of each rodent. This CT-derived weight (obtained by multiplying the total volume of AT, LT, and ST by the representative tissue density) can be compared with the gravimetric weight, measured at the time of imaging; this approach provides an additional estimate of the accuracy of CT-derived body composition.
MATERIALS AND METHODS
Candidate animals for our μCT-based whole body composition analysis were chosen from two preexisting animal studies that had received prior ethics approval from the Animal Use Subcommittee of the Council on Animal Care at the University of Western Ontario (ethics protocol nos. 2008-095 and 2006-129) in accordance with the Guide for the Care and Use of Laboratory Animals, as approved by the Council of the American Physiological Society. Group A, the small-animal osteoporosis disease model and treatment study, consisted of 25 female Sprague-Dawley rats obtained from the Charles River Laboratory (Wilmington, MA) at 3 mo of age with an average weight of 270 g. They were given free access to water and regular rat chow. Group A was subdivided into four groups: group A-1 (5 rats) were sham-operated controls, receiving no treatment; group A-2 (3 rats) were weight controls, also receiving no treatment; group A-3 (5 rats) underwent ovariectomy with treatment A; and group A-4 (12 rats) underwent ovariectomy with treatment B. Treatment A consisted of phosphate-buffered saline, and treatment B consisted of hyaluronic acid, (1 mg/kg given by gavage), both of which were administered each morning (from Monday to Friday) through gavaging. After 9 wk, group A animals were euthanized in a CO2 chamber and immediately frozen to −20°C until imaged using μCT. All animals were grouped together for the purposes of this study.
Group B, the small-animal diabetic disease model and treatment, consisted of 52 male Sprague-Dawley rats. The experiment ran for 8 wk. Rats were subdivided into six groups. Groups B-1–B-3 underwent varying aerobic exercise regimens and multiple, low-dose, intraperitoneal injections of streptozotocin (1 injection/day for 5 consecutive days, totaling 5 injections/animal at 20 mg·kg−1·injection−1) to mimic type 1 diabetes (26). Groups B-4–B-6 underwent varying aerobic exercise regimens only. At the end of 8 wk, group B rats were scanned in vivo using μCT and then euthanized for further analysis.
Group A (25 postmortem female) and group B (52 live male) Sprague-Dawley rats were scanned using a cone beam, volumetric, μCT scanner (GE eXplore Locus Ultra). At the time of imaging, rats were placed on a digital balance and the weights recorded. The live rats in group B were anesthetized through an intraperitoneal injection consisting of a mixture of ketamine (75 mg/kg) and xylaxine (10 mg/kg) during the scan interval, which was typically <5 min. Scans were acquired using a 16-s, 1,000-view protocol, with an approximate transverse field of view (FOV) of 15.4 × 15.4 cm, longitudinal extent of 10.2 cm, X-ray tube potential of 120 kVp, tube current of 20 mA, and no additional filtration. Because the length of each rat was greater than the longitudinal extent of the FOV, three image volumes were acquired, each longitudinally displaced by 90 mm from the previous scan, to cover the entire rat contiguously from head to tail, a total extended FOV length of 27 cm. Following 3D cone beam filtered back-projection reconstruction, these three volumetric segments were digitally “stitched” together to provide a single 3D volume, which contained the full longitudinal extent of each rat.
To assess the precision of our scanner to perform body composition measurements, a subset of six animals from group A (the postmortem female rat group) was imaged repeatedly. Each of these six animals was imaged six times (repositioning the scanner bed between each scan) for a total of 36 whole body scans and 108 image volume segments (i.e., 3 sections × 6 rats × 6 repeated measurements = 108 acquired volumes). It has been estimated previously by Gluer et al. (11) that six repeated measurements of bone mineral density (equivalently BMC) on six different subjects will result in an appropriate 95% confidence in a calculated precision value.
Included in each μCT scan were several cylindrical calibration samples containing known increasing amounts of bone mineral equivalent densities. The first cylinder was made of a tissue equivalent plastic with zero BMC, whereas the last cylinder was an epoxy-based, cortical bone-mimicking calibrator (SB3; Gamex RMI, Middleton, WI) having a bone mineral equivalent of 1,100 mg/cm3 (36). To avoid scanner drift, a water phantom was also included in each image to ensure accurate calibration of CT grayscale intensities in Hounsfield units (HU) for each imaged volume.
The total radiation dose to the live rats, consisting of three scans each at 120 kVp (20 mA, 16 s), was measured from X-ray ionization chamber measurements. To perform the exposure measurements, the gantry of the scanner was held at a single angular position, and an ionization chamber (model 96035; Keithley, Cleveland, OH) was placed in the center of the FOV, perpendicular to the imaging plane, and contained completely within the X-ray beam. The ionization chamber was connected to a dosimeter (model 35617; Keithley) and the electrical charge (in nC) recorded by the ion chamber for a 1.1-s exposure at 90 mA, 120 kVp. Using the model-specific ion chamber conversion factor of 0.2327 R/nC, the experimental exposure in Roentgens per mAs was calculated. The total exposure for our 16-s, 20-mA scan at 120 kVp was multiplied by the experimentally measured exposure conversion factor. The total absorbed dose (in air) for a single scan was then calculated using a conversion factor of Dair = 0.876 cGy R−1 (33).
Images were reconstructed using the Feldkamp cone beam filtered back-projection algorithm at twice the detector pixel spacing, resulting in a reconstructed, isotropic voxel spacing of 308 μm (8). The three acquired volumetric images were cropped (to remove cone beam boundary artefacts) and digitally stitched together to produce a complete, volumetric image of the entire rat, having a total image size of 486 × 160 × 880 voxels in axial planes x, y, and z, respectively. All reconstructed images were analyzed or processed using custom application software in C + + and compiled using g + + (version 4.0.2; Free Software Foundation, Boston, MA).
To evaluate the accuracy of whole body composition from reconstructed image volumes, at the completion of the small-animal diabetic disease model study (group B), the animals were euthanized and all of the muscles composing the lower right leg between the knee and the ankle were excised and gravimetrically weighed. For each rat, the total mass of the muscles between the knee and the ankle, the soleus, the plataris, the gastricnemius, the tibialis anterior, and the extensor digitorum longus, were combined to provide a measured lean tissue mass in a defined region within the rat. Carefully sculpted regions of interest (ROI) in the corresponding reconstructed image regions, containing all tissues present between the knee and the ankle in the right leg, were used to determine the CT-derived lean tissue mass and compared with the gravimetrically measured lean tissue mass.
Calculation of adipose and lean tissue volumes.
Single-energy CT is capable of segmenting body composition into three tissue types on the basis of the grayscale signal intensity within each voxel, so long as the following conditions are met: 1) the tissues are anatomically compartmentalized into homogeneous regions that are relatively large (in comparison with the reconstructed voxel size), and 2) the tissues differ sufficiently in linear attenuation coefficient such that they can be separated successfully, even in the presence of random noise in the CT image. We have implemented whole body CT imaging in rodents, with sufficiently small voxels and reduced image noise, so that these conditions will be met. For each reconstructed volume containing a whole rat specimen, a total of three global grayscale thresholds were derived and used to characterize voxels as AT, LT, or ST. The three thresholds were chosen using an analysis of volumetric histograms from a subset group that was representative of the larger group. In one of the image volumes for each of the six repeatedly imaged rat specimens, a 3D ROI (25 × 25 × 25 voxels in size) was placed in a region containing visible portions of AT, LT, and ST. A grayscale histogram was generated for each of the six ROIs. An averaged histogram of the six ROIs was calculated, and two peaks (corresponding to AT and LT) were identified. Based on the average histogram, the threshold separating the AT and LT peaks (CTAT/LT) was determined using the method of Otsu (24). A Gaussian distribution was fit to the AT peak and another to the LT peak. A threshold, separating air and AT (CTAIR), was chosen at three standard deviations below the mean of the AT peak. Another threshold, separating lean tissue from skeletal tissue (CTST), was chosen at three standard deviations above the mean of the LT peak.
After application of the derived thresholds, voxels labeled as either AT, LT, or ST were visually inspected to confirm that the appropriate voxels were correctly identified.
The AT volume (ATV), LT volume (LTV), and ST volume (STV) were calculated as (1) (2) (3) where Δx is the linear dimension of an isotropic voxel (i.e., 0.0308 cm in our study).
In calculating a ST volume, voxels classified as ST are considered to consist of an inorganic component (hydroxyapatite) and an organic component (collagen and fluid).
Calculation of BMC.
The whole body BMC for each Sprague-Dawley rat was calculated on the basis of the empirical relationship between known amounts of mineral content and CT number, using calibrators placed within each μCT image volume. Two cylindrical ROIs containing ∼1,000 voxels or 30.0 mm3 were placed within each image volume, one in the tissue-equivalent (TE) calibrator and the other in the SB3 calibrator. The average CT value within each calibrator was calculated and represented in HU as CTTE and CTSB3. All voxels having intensities above the bone threshold (>CTST) are processed to calculate a total whole body BMC as (4) where 1,100 is the known mineral value of SB3 in mg/cm3 and NST represents the total number of ST voxels.
The histograms from the six ROIs placed in each of the six postmortem rat specimens and the averaged histogram can be seen in Fig. 1, A and B. The threshold separating the AT and LT peaks, determined by the Otsu method (24), was found to be CTAT/LT = 5 HU. A Gaussian distribution was fit to the AT peak, having a mean at −81 HU, a SD = 27 HU, and a goodness of fit parameter of r2 = 0.96. A second Gaussian distribution was fit to the LT peak, having a mean at 89 HU, a SD = 22 HU, and a goodness of fit parameter of r2 = 0.99. Therefore, the threshold separating AT and air at 3 SD below the AT peak was found to be CTAIR = −186 HU, and the threshold separating LT and ST at 3 SD above the LT peak was found to be CTST = 155 HU. Therefore, the thresholds CTAIR, CTAT/LT, and CTST used in the tissue volume calculations in the 52 live male and 25 postmortem female rat specimens were −186, 5, and 155 HU, respectively. The result of this classification scheme is illustrated in one of the imaged live male rat specimens in Fig. 2.
For all rats, their total whole body ATV, LTV, and STV were calculated and the average volumes within each group reported in Table 1.
For all voxels classified as ST, the whole body BMC was calculated as described in Eq. 4, using the experimentally observed CT numbers for the TE calibrator (CTTE = 75 HU) and SB3 cortical bone equivalent (CTSB3 = 1,930 HU). Shown in Table 2 is the measured weight and the average whole body BMC for each of the six postmortem female rats, which were repeatedly measured six times. These repeated BMC measurements are shown in Fig. 4.
The precision of the whole body BMC and tissue volume measurements.
The experimental precision of the scanner to report whole body BMC, ATV, LTV, and STV from μCT images of six adult postmortem Sprague-Dawley rats was calculated as the root mean squared average of each repeated-measures variance. These values were then divided by the mean value of each measurement to report a final experimental precision value as a percentage. The calculated experimental precision of the μCT system to report whole body values for BMC, ATV, LTV, and STV was found to be ±0.49, 1.9, 1.7, and 0.57%, respectively, of the average mean value. Each repeated measurement and the average of that measurement can be seen in Figs. 3 and 4.
An assessment of the accuracy of the tissue mass measurements.
The accuracy of our whole body composition method was evaluated by comparing known, excised, and gravimetrically weighed measurements of muscle tissue (lean tissue) in the right lower leg of the male group of rats to the image-calculated lean tissue measurement within the μCT image region containing the same muscular compartments. A paired t-test between columns of the gravimetrically weighed muscles vs. the CT-derived lean tissue muscle weights for each individual rat showed no significant difference (P = 0.134). Table 3 lists the total averaged lean tissue mass from the postmortem gravimetric and CT-derived analysis that was found to be 3.5 ± 0.1 and 3.6 ± 0.1 g (SE), respectively. The percent difference between the average experimentally measured lean tissue masses vs. the image-derived masses was 2.2%. An example of the user-defined ROI containing all of the muscular compartments of the lower right leg and the resulting identified lean tissue voxels within that ROI can be seen in Fig. 5.
The total whole body dose was calculated to be 10 cGy (delivered by the 3 displaced 16-s scans at 120 kVp, 20 mA); this value is ∼2% of the reported LD50/30 for small animals such as rats (17) or mice (25, 30).
Comparison of tissue masses to gravimetrically measured masses.
The conversion of AT, LT, and ST volumes into tissue weights was performed by multiplying each volume by the respective representative tissue densities 0.95, 1.05, and 1.92 g/cm3, as listed by the international commission of radiological units (11a). The cumulative CT-derived tissue weights were calculated and found to be 362 g for the average of the postmortem female group of rats and 431 g for the average of the live male group of rats. These values differed from their average gravimetrically measured group by 2.5 and 1.1%, respectively.
Whole body composition results.
The average whole body CT-derived tissue masses of rats using μCT images of whole rat specimen, as well as the whole body BMC for the postmortem female and live male groups, are listed in Table 4.
We have demonstrated that we can successfully quantify adipose, lean, and skeletal tissue in adult Sprague-Dawley rats using conventional single-energy μCT images reconstructed with 308 μm of isotropic voxel spacing. By imaging entire animals, we were able to calculate the total whole body contribution of the adipose, lean, and skeletal tissue volumes, and subsequently the adipose, lean, and skeletal tissue masses, using representative tissue density values.
A total of 25 postmortem female and 52 live male Sprague-Dawley rats were scanned, and their total whole body ATV, LTV, and STV were calculated. From these values, the total tissue masses were tabulated. The average CT-derived total body weights were 362 g for the postmortem female group and 431 g for the live male group, which differed from the average gravimetrically measured weight by 2.5 and 1.1%, respectively.
These values demonstrate that our whole body composition technique has excellent performance for a range of animal models that may express quite different body compositions; i.e., in our study, postmortem osteoporosis model female rats reported 20% of their total body composition to consist of adipose tissue, whereas the live diabetic model male rats expressed only 8% of their total body composition to consist of adipose tissue. Through repeated measurements, the precision of the μCT system's ability to report BMC, ATV, LTV, and STV was found to be ±0.49, 1.9, 1.7, and 0.57%, respectively, of the average mean value. This indicates that our imaging system is highly sensitive in monitoring changes in BMC and STV and slightly less sensitive in comparison with changes in AT and LT volumes, as one should expect due to the similarity of CT numbers between the two tissue types.
We investigated the accuracy of our technique by comparing excised muscle tissue measurements (i.e., lean tissue mass) to the image-calculated lean tissue mass measurement of the same muscular components within a user-defined image region that differed on averaged by 2.2%. This suggests that our technique is highly accurate even in localized regions of interest. The slight overestimation of our image-derived lean tissue mass could be due to the marrow component that was included as lean tissue or to loss of blood from the excised muscle tissue that may have been considered lean tissue in the CT-derived lean tissue mass of the animal when imaged in vivo. Understandably, this is a singular measurement of the accuracy of our whole body composition technique and does a represent a comprehensive analysis on the accuracy of our technique, which would require lengthy destructive chemical analysis and is beyond the scope of our study. However, Cortright et al. (6) report values derived from chemical analysis for the total bone ash (i.e., total whole body bone mineral content) percent by weight of 2.9–3.8% that compare favorably with our findings of 2.7–2.9%. Any further comparison would be speculative because chemical analysis does not directly report lean tissue mass, and there exist large observational differences in adipose tissue masses (i.e., fat) even within each study, depending on a given animal's environmental factors (i.e., nutrition, frequency of exercise, sex difference, individual variability).
Further improvements to the accuracy of our technique would result from an improved estimation of the average density value of skeletal tissue that is defined in this article as to consist of an inorganic component (hydroxyapatite) and an organic component (collagen and fluid). In this article, we assigned the density of skeletal tissue to be that of cortical bone 1.92 g/cm3, and this is likely an overestimation that results in a systematic overestimation in the mass of skeletal tissue. However, since skeletal tissue voxels represent only 6% (Table 1) of the total voxels in whole body μCT scan of rats, a large difference in the density value for skeletal bone does not result in a large difference in total body mass. In addition, a more accurate calculation of the skeletal tissue mass could be implemented at any time when an improved value for skeletal tissue becomes available but this was not found to be critically necessary for the purposes of this article.
A broad range of applications could benefit from simple, fast, nondestructive, whole body composition analysis of small animals. Numerous epidemiological studies related to diabetes (26), alcohol abuse (20), or high-fat diets (2, 4, 5) that observe and monitor whole body compositional changes in small animals would be ideal candidates. Whole body compositional analysis could also be a valuable tool for high-throughput characterizations of phenotype expression in small animals (16) or in the understanding of disease processes that affect more than one tissue in unexpected ways, such as tumor treatment.
The ability to efficiently perform whole body compositional analysis in rats relies on the use of a rapid, volumetric, cone beam μCT scanner, which is similar to a clinical CT scanner in that a live subject can rest on the scanner bed, access to the subject is maintained, and a slip-ring gantry rotates around the subject. Despite the fact that this new in vivo technology may not presently be widely accessible in many small-animal research facilities, we believe that it will inevitably be adopted more frequently because of its ability to acquire very high-resolution image data (>1 GB/volume) quickly and efficiently. We have also shown that whole body μCT imaging of rats can be performed at an acceptable dose level of 10 cGy. This exposure to ionizing radiation should be safe for small-animal longitudinal studies, given that previous studies have indicated that the repair processes in rodents will allow them to neutralize 25 cGy of radiation/day (23).
Our in vivo image-based whole body compositional analysis of small animals has been limited to rats, but it could be applied equally to other small animals. In our study, μCT images were reconstructed with detector elements binned 2 × 2, resulting in an isotropic voxel size of 308 μm. This voxel size was found to be sufficiently small enough as to avoid partial voluming effects between different tissue compartments but large enough that the whole body image volume files were manageable (∼1 GB). If the μCT images were reconstructed at the nominal resolution, this would have resulted in half-isotropic voxel size (154 μm) but eight times the file size. For other animals, the appropriate voxel size would have to be determined a priori, but if the voxel size were scaled according to animal anatomy, voxel sizes between 30 and 60 μm would be desired for mice. For larger animals such as rabbits, large image volumes may make routine whole body composition analysis cumbersome to acquire and manage. It should also be mentioned that our whole body composition technique requires μCT image volumes of small animals to be generally free of image artifacts such as “image blurring” due to motion or “image streaking” due to beam hardening around highly attenuating components. For live animals, if they are adequately sedated, any associated body composition errors due to motion taken in the context of the whole animal body composition appear negligible.
This study verifies the capability of single-energy μCT scans to quantify whole body composition of small animals. This high-throughput technique will have many applications in longitudinal studies of rats and mice, providing the ability to detect changes in body composition of only a few percent.
Financial support was provided by the Canadian Institutes of Health Research (MOP-89852 and CCT-83029), the Ontario Research and Development Challenge Fund, and Cogent Solutions. D. W. Holdsworth holds the Dr. Sandy Kirkley Chair in Musculoskeletal Research at the Schulich School of Medicine and Dentistry, and E. A. Turley is supported by the Breast Cancer Society of Canada.
No conflicts of interest, financial or otherwise, are declared by the authors.
We thank Anu Bhalla for assistance with the animal experiments.
- Copyright © 2010 the American Physiological Society