Main

Precise noninvasive measurement of body composition is of great value to characterize quality of growth and nutritional status in premature and term infants under healthy and pathologic conditions (1). Besides classic balance studies, direct assessment of the accretion of protein and fat mass will help to evaluate the impact of nutritional or therapeutic regimens in interventional studies. Two modern methods seem to be promising to assess noninvasively and in vivo body components of newborn infants: DXA and MR using either MRI or MRS sequences. The main aim of the current animal study was to validate body composition measurements by DXA, MRI, and MRS relative to chemical carcass analysis as the “gold standard.”

The principle of DXA has been described previously (2–4). Using whole-body x-ray scans obtained at two different energy levels, body composition is calculated by the differential absorption of each pixel using an acrylic step-phantom for calibration. In adults, DXA has been cross-validated against other methods, including chemical carcass analysis, and was found to give reliable estimates of adult body composition (5–13). In neonates, DXA has already been used in a number of studies (14–23). However, validation data for this population obtained by chemical carcass analysis of piglets are somewhat contradictory (17, 18, 21, 22). This may be related to incomplete recovery or systematic errors during carcass preparation (e.g. fat adhesion to grinder walls, inhomogeneous fat distribution in the homogenate). It was therefore one aim of the current animal study to minimize systematic errors during carcass preparation and to compare data on carcass analysis with DXA data. A second aim was to develop a reduced-size step-phantom to save scanning time and to compare the measurements with the classic adult-size phantom.

The major advantage of MR techniques for measuring neonatal body composition is the lack of radiation exposure. MR basically provides two ways to estimate body fat content, using either imaging or spectroscopy sequences. The classic imaging approach allows tissue differentiation because MR relaxation parameters are tissue-dependent and lead to different signal intensities. Calculation of body composition may be performed by means of volumetry using either tissue segmentation algorithms or point-counting methods. Classic MRI has been successfully used in animals and adults (24–29) and already applied to assess fetal growth and nutritional status (30–33). In a recent study, assessment of adipose tissue in infants up to 3 mo of age using MRI volumetry was charged with considerable interobserver bias (34), and to the best of our knowledge, no study has been published in which MRI was validated to assess body composition during the early postnatal period.

Alternatively, fat and water content may be assessed by proton MRS (1H-MRS) because the resonance frequencies of water and fat are slightly different by approximately 3.5 ppm (parts per million). To avoid distortions and overlap of both resonances, the magnetic field must be extremely homogeneous within the investigated volume. Because of physical limitations, these homogeneity criteria may only be achieved in volumes of a few hundred milliliters, but not in large bodies (35, 36). However, it is possible to overcome these limitations by CSI (37, 38). CSI applies additional phase-encoding steps for spatial encoding, such that the body is divided into a number of voxels. Although different voxels experience different magnetic field strengths, the magnetic field is relatively homogeneous within one specific voxel, resulting in a separated registration of water and fat signals in each voxel. The signals of different voxels can then be corrected and summed up. 2D-CSI has already been used to measure fat/water ratios of small animals (i.e. rats, chickens), but 2D-CSI encodes for only two spatial directions, making it necessary to measure the whole body slice by slice, which is time-consuming (39, 40). The last aim of this study was to adapt and evaluate a 3D-CSI sequence using space-encoding steps for all three spatial dimensions. This should allow simultaneous measurement of the whole body without adjusting homogeneity of the magnetic field slice by slice.

METHODS

Animals and experimental procedures.

The study was approved by the local commission for animal experiments. Twenty-three piglets (Landrace breed) were investigated, covering an age and weight range from 1 to 37 d and from 848 to 7550 g, respectively. On each study day, the animal was weighed on an electronic scale, and total body water was determined using deuterium dilution. The animal was then killed, and within 2 h, body composition was assessed using MR and DXA. The body was stored overnight at 4°C. The next morning, the carcass was homogenized, and the samples were stored at −80°C until chemical analysis was performed.

Total body water measurements.

After taking a predose blood sample, 0.5 g of 99.8% deuterium oxide per kilogram BW was applied by a syringe and a nasogastric feeding tube. Analysis of 3-h postdose deuterium enrichment was performed in duplicate using Fourier-transformed infrared spectroscopy as previously described (41, 42). The effect of isotope sequestration was corrected for using a factor of 1.04 (43, 44). Owing to technical problems with tracer administration, body water was successfully measured only in the last 15 animals (numbers 9–23).

DXA measurements.

After killing the animal by intracardial injection of barbituric acid, DXA scans (QDR-1500, Hologic, Waltham, MA) were performed in triplicate, without repositioning the animal, using the pediatric whole-body scanning procedure and the pediatric pad as well as the adult tissue bar for calibration (serial no. 702). Scans were analyzed using a modified infant software [version 5.67, same version as used in (18) and (21)] with separate drift corrections for both x-ray energy levels. The modification was implemented by the manufacturer as a result of a pilot study, in which we investigated the performance of our DXA system for the measurement of small bodies (data not published).

A baby step-phantom to reduce scanning time was created by cutting a second tissue bar (serial no. 1036) to one fourth of its original size (50% of width, 50% of length). In 11 piglets, additional DXA scans were performed using this modified bar.

MRI and 3D-CSI.

In 21 piglets, sagittal T1-weighted spin-echo images were obtained using a 1.5-T whole-body MR system with the in-built body coil (SIGNA, General Electric, Milwaukee, WI). No part of the body exceeded the homogeneous volume, and the body axis was along the magnetic field to ensure optimal homogeneity of the magnetic and radio-frequency fields for all parts of the body [repetition time, 340 ms; echo time, 16 ms; 4 excitations per phase step; matrix, 256 × 192; slice thickness, 5 mm; interslice distance, 0 mm; field of view adjusted to piglet size; Cavalieri principle using random shift of starting position within the range of one slice thickness to avoid biased volumetry in objects with heterogeneous composition (45)]. The whole set of images was analyzed for body volume and composition using the point-counting method (45). According to the anatomic structures and signal intensities, pixels were assigned to one of the following body compartments: s.c. fat, visceral fat, bone, vessel, or lean mass. Ambiguous pixels suffering from a clear discrimination between visceral fat and lean mass were assigned in equal parts to each of the two compartments.

After MRI, 1H-MRS (3D-CSI) was performed without changing the animal's position (37, 38). The 3D-CSI MRS sequence was not available in the first four animals; the measurements were therefore performed in the last 19 of the 23 piglets (numbers 4–23). Basal shim values were used, and radio-frequency power adjustment was taken from the imaging prescan. The CSI pulse sequence used the signal of the free-induction decay and had a repetition time of 300 ms, acquiring two responses per phase-encoding step, and an overall scanning time of 20.4 min. Owing to the time needed for phase encoding, data acquisition started 0.6 ms after radio-frequency excitation. Voxel size was between 5.0 and 28 mL.

Data processing included apodization with an 8-Hz exponential window and zero-filling to 1024 points in the time dimension, symmetric Fermi apodization (diameter, 65%; transition width, 25% of total data width) for all spatial dimensions, and zero-filled, resulting in a 1024 × 16 × 16 × 16-point data set in the first spatial dimension. A four-dimensional fast Fourier transformation revealed a CSI data set with 1024 chemical shift points (512 each real and imaginary points) and 16 × 16 × 16 spatial points. Each of the resulting spectra was frequency corrected and then summed up (user-defined routine written in IDL, Research Systems, Boulder, CO). The resulting spectrum was corrected for phase and baseline and then integrated using a previously published procedure (46, 47). Total time for data processing was 5 min per fit.

Saturation factors of water and fat were measured in a representative subset of four piglets by two spatially unresolved spectra at 300 ms and 16 s to correct MR signal intensities (1.0 and 0.55 for water and fat, respectively).

To calculate the total amount of water and fat, MR signal intensities were corrected for different proton densities of water and triglycerides. A factor of 9.00 was used for water, the factor for triglycerides was 8.42: mean content of 100.4 hydrogen ions per molecule of triglycerides (mean molar weight, 846.0). Data about triglyceride composition in adipose tissue of piglets (age range, 1–35 d) were given by Le Dividich et al. (48) .

Absolute fat content by MRS was calculated from fat-to-water ratios using in vivo body water determination by deuterium dilution and was compared with crude fat as measured by chemical analysis.

Homogenization procedure and chemical analysis.

The homogenization protocol was designed and tested for complete recovery of carcass material as well as for perfect homogenization to assure that aliquots subsequently measured by chemical analysis were representative for total carcass composition. After measuring BW, the carcass was passed through a butchery grinder (Talsa, type P114 PU5, J. Koch, Malters, Switzerland) equipped with a set of three rotating cross-formed knifes and two rotating knife plates having bores with diameters of 20 and 8 mm (Lico, nos. Z020.L4104, L2104, and L3408, J. Koch). The grinder was rinsed with water, the exact rinsing volume being known from weighing water containers before and after use (mean, 1410 ± 190 mL). The homogenized moiety and all of the eluate passing through the grinder were caught in metal bowls, whose weight was measured and immediately frozen at −30°C. From the difference of expected weights (carcass plus rinsing water used), the mean residual volume of the eluate in the grinder was calculated to be 118 ± 31 mL. The result of analysis of solid substances in this residual eluate was 0.2 ± 0.1 g. Mean recovery of carcass mass after homogenization therefore exceeded 99.0%.

The homogenate was lyophilized in a standard device (Christ, Delta 1–24/K, Kuehner AG, Birsfelden, Switzerland). The resulting dry matter was passed through a second grinder using knife plates with bores of 1.0 and 4.0 mm (Brabender, Wiley System, Stiner AG, Kilchberg, Switzerland). The quality of the homogenization procedure was assessed by measuring contents of crude protein, ash, and residual water in 10 aliquots of one homogenate that was chosen randomly. Coefficients of variations were 1.7, 2.1, and 1.8%, respectively, and compared well with the analytic error of the methods used. It may therefore be concluded that complete homogenization was achieved by the procedure used.

Chemical analysis (duplicate measurements) was performed in the routine laboratory of the Swiss Federal Research Station for Animal Production, Posieux, a certified reference laboratory for analysis of animal feeds. Total ash was measured gravimetrically after combustion at 550°C for 4 h (CV < 2.0%), contents of calcium and inorganic phosphorus of residual ash were measured photometrically (CV < 2.0% and <1.0%, respectively), crude fat was measured according to Berntrop (hydrochloric acid for protein denaturation) and Soxhlet fat extraction (CV < 2.5%), crude protein was measured according to Kjeldahl using photometric determination (CV < 0.8%), residual water of the lyophilizate was measured gravimetrically after heating at 105°C for a period of 3 h (CV < 1.0%), and gross energy content was measured in an adiabatic bomb calorimeter (CV < 0.6%) and was additionally calculated from crude fat and crude protein assuming an energy content of 36.6 MJ/kg crude fat and 24.2 MJ/kg crude protein (49). Body water was calculated as carcass weight minus total weight of the lyophilizate plus residual water content of the lyophilizate. Fat-free mass was calculated as carcass weight minus crude fat.

Statistical analysis.

Data are presented as mean ± SD. Standard statistical methods (descriptive statistics, correlation, and regression analysis) were used to compare DXA measurements with chemical analyses. Paired t test was used to compare data obtained by the two step-phantoms. Conversion equations were developed by taking the inverse of the regression equation, the systematic error (i.e. accuracy) was assessed by calculating the mean of the residuals, and the random error (i.e. precision) was calculated as the SD of the residual sum of squares. Levels of significance were set to p< 0.05.

RESULTS

Body composition by chemical analysis.

Table 1 shows body composition of the 23 piglets as determined by chemical carcass analysis. Fat content varied from 7 to 674 g (0.7–13.1% body fat). Protein content accounted for 139 ± 9 g/kg BW (range, 120–154 g/kg BW). Composition of fat-free mass (935 ± 37 g/kg BW) is presented in Table 2, thereby assuming that fat mass is almost free of water and protein. Mean water content was 804 ± 11 g/kg fat-free mass (range, 785–825 g/kg fat-free mass, r2 = 0.999). Hydration of fat-free mass was dependent on BW and decreased with increasing BW (Fig. 1, r2 = 0.419). There was no significant correlation between hydration of fat-free mass and postnatal age (r2 = 0.082).

Table 1 Body composition of 23 piglets as measured by chemical analysis
Full size table
Table 2 Composition of fat-free mass in 23 piglets as measured by chemical analysis
Full size table
Figure 1
figure 1

Relationship of water content of fat-free mass vs BW.

Full size image

Correlation of body composition as measured by DXA and chemicalanalysis.

CV for the repeated measurement of BMC, FM, fat-free mass, and BW were 1.0, 4.6, 0.4, and 0.06%, respectively. Correlation of BW as measured by electronic scale and by chemical analysis was high (r2 = 0.999), but DXA overestimated BW by a factor of 1.2%. DXA BMC was highly correlated with chemically determined contents of total ash, calcium (Fig. 2, top), and phosphorus (r2 = 0.993, 0.992, and 0.992, respectively). Calcium and phosphorus accounted for only 44.1 ± 4.2% and 29.1 ± 1.8% of DXA BMC. Total ash exceeded DXA BMC by a factor of 1.72 ± 0.13. Figure 3 (top) shows that DXA FM was highly correlated with total fat by chemical analysis (r2 = 0.961), but the regression equation was different from the line of identity (slope = 0.763, intercept = 52.4 g). DXA lean mass was highly correlated with fat-free mass and with total body content of crude protein (r2 = 0.999, slope = 1.012, intercept = −77.7 for fat-free mass;r2 = 0.982, slope = 6.26, intercept = 107.3 for crude protein).

Figure 2
figure 2

Top, Correlation of DXA whole-body BMC vs chemically determined calcium content. Bottom, Plot of residuals after conversion.

Full size image
Figure 3
figure 3

Top, Correlation of DXA FM vs crude fat by chemical analysis. Solid line, regression line;dotted line, line of identity. Bottom, Plot of residuals after correction.

Full size image

Conversion and correction equations.

The high correlations of DXA and chemical analysis for the measurement of calcium, phosphorus, total ash, and fat justify the development of conversion equations. EQUATION 1 EQUATION 2 EQUATION 3 EQUATION 4 and EQUATION 5

Mean difference between corrected DXA and scale body weight (i.e. accuracy) was −0.2 g with a random error (i.e. precision) of ±9.5 g, a statistically not significant difference. Values before conversion were 17.8 ± 24.7 g (p< 0.01).

Fig. 2, bottom, shows a mean difference between converted DXA BMC and chemically determined calcium content of 0.0 g with a random error of ±1.9 g. The difference before conversion was 38.6 ± 22.7 g (p< 0.001).

Figure 3, bottom, shows the plot of the residuals for DXA FM after correction: the mean difference between DXA and crude fat was −11.3 ± 61.1 g before and 0.0 ± 43.7 g after correction. The relative error for fat mass was 10% over the whole range studied except for the first data points (absolute fat content of 7 and 10 g, relative error of 100 and 200%).

Measurement of gross energy content.

Gross energy content as calculated from crude fat and crude protein correlated well with gross energy content measured by bomb calorimetry (r2 = 0.999), but was slightly underestimated by a factor of 2.8% (Fig. 4, top). The corrected DXA data for FM and fat-free mass were used to calculate gross energy content. Protein content of fat-free mass was set to 149 g/kg fat-free mass (Table 2). Figure 4, bottom, shows that gross energy content calculated from converted DXA data are highly correlated with direct measurement by bomb calorimetry (r2 = 0.984).

Figure 4
figure 4

Top, Correlation of gross energy content measured by bomb calorimetry vs that calculated from crude fat and crude protein by chemical analysis. Bottom, Gross energy content directly measured by bomb calorimetry and calculated from the corrected DXA FM and fat-free mass.

Full size image

Measurements using the baby step-phantom.

Analysis of tissue bar absorption data revealed 20 data points for each calibration step for the small tissue bar (serial no. 1036) compared with 100 data points for the original one (serial no. 702). These 20 data points were sufficient to establish a precise calibration line. Calibration parameters obtained for the small tissue bar were not statistically different from those obtained for the original phantom (paired t test).

Body composition data obtained with the small phantom were highly correlated with the results by the original (r2 = 0.999 for BMC, 0.991 for FM, 0.999 for fat-free mass and BW). Mean differences for BMC, FM, fat-free mass, and BW were 1.1 ± 1.6, −11.9 ± 23.7, 16.1 ± 26.7, and 5.6 ± 8.2 g, respectively, and were not statistically different from zero (paired t test).

Measurement of total fat using MRI versus chemicalanalysis.

Figure 5, top, depicts the relationship of total fat as measured by MRI and chemical analysis. Although the data obtained by both methods are highly correlated (r2 = 0.923), MRI gives systematically higher body fat content than chemical analysis by a factor of 2.5, with a zero intercept of 57 mL. Figure 5, bottom, shows the correlation of total body volume with body weight (r2 = 0.990). The slope of the regression is 0.910 mL/g; the intercept is −37.0 mL. Mean specific carcass gravity was 1.080 ± 0.032 g/mL.

Figure 5
figure 5

Top, Correlation of body fat determined by MRI and chemical analysis. Solid line, regression line;dotted line, line of identity. Bottom, Correlation of MRI body volume vs body weight.

Full size image

Measurement of total fat using MRS versus chemicalanalysis.

Figure 6 illustrates the effect of postprocessing on the whole-body spectrum. The raw spectrum (top trace) was obtained by summing up all spectra without correction. In the second trace, the spectra were individually corrected for frequency shifts before being summed up, which significantly improves the separation of lipid and water resonances. The third trace shows the fitted spectrum. The bottom trace represents the residuals.

Figure 6
figure 6

Effect of correction algorithm. Top, Addition of all spectra without correction. Resonances of fat and water largely overlap. Top middle, Addition of all spectra after individual frequency correction, resonances of fat and water are clearly separated. Bottom middle, Fitted spectrum. Bottom, Plot of residuals.

Full size image

Figure 7, top, shows the relationship of fat-to-water ratios as measured by 3D-CSI and chemical analysis in 19 piglets. Both methods were highly correlated (r2 = 0.984); however, fat-to-water ratios determined by MRS were systematically lower by a factor of 0.772 with a negligibly small zero-intercept of 0.003. Figure 7, bottom, shows the relationship of absolute body fat by MRS versus crude fat when calculated from the MRS fat-to-water ratio using deuterium dilution, which was successfully done in 15 piglets. The correlation of both methods was high (r2 = 0.988); again, MRS showed a slightly smaller amount of body fat (factor of 0.841) with a negligibly small intercept of 8.6 g.

Figure 7
figure 7

Top, Correlation of whole body fat-to-water ratios measured by 3D-CSI vs chemical analysis (n= 19). Bottom, Correlation of total body fat by 3D-CSI and deuterium dilution vs chemical analysis (n= 15). Solid lines, regression lines;dotted lines, lines of identity.

Full size image

DISCUSSION

General aspects.

In this study we have validated DXA, MRI, and MRS against chemical carcass analysis. Piglets with a BW of 0.9–7.6 kg were chosen to simulate conditions found in term and preterm newborn infants from birth up to the early postnatal period of 6 mo. Complete carcass recovery and perfect homogenization of the resulting dry matter was achieved, which we consider a necessary prerequisite for acceptance of chemical analysis as gold standard. We were able to show that DXA and MRS measurements were highly correlated with chemical analysis. We conclude, therefore, that DXA and MRS give accurate and precise estimates of neonatal body composition.

DXA versus chemical analysis.

Currently, only a few studies have been published in which neonatal DXA measurements were validated against chemical analysis of the whole carcass (17, 18, 21, 22). The DXA devices used in these studies were similar to the one we used, but some applied older software versions (17, 22).

In the first study, using the older pediatric whole-body software (version 6.01), DXA BMC, FM, and fat-free mass were found to correlate poorly with chemical analysis (r= 0.16, 0.06, and 0.92, respectively) (17). Reanalysis using the actual infant whole-body software version 5.56 led to an improved correlation, which, however, still suffered from considerable individual errors, making its routine use debatable (22). However, in our opinion, there was evidence that carcass recovery after homogenization might not have been complete. In contrast to the classic rules of arithmetic, the sum of the mean compartment sizes did not equal the mean of BW (mean differences of 30 and 147 g in the 1.6 and 6.0 kg groups, respectively), which indicates that some material must have been lost during carcass preparation and might have led to a systematic error. The poor correlation found in both studies might therefore in part be explained if it is assumed that preferentially fat was lost because of its high affinity to metal walls.

In contrast, another validation study in piglets (n= 13; weight, 1470–5510 g) revealed that DXA measurements were highly correlated with chemical analysis, indicating for the first time that DXA precisely reflects chemically determined carcass components (r= 0.992 and 0.971 for BMC and fat mass, respectively), but needed to be corrected by conversion equations (21).

The results of our study support the data given in a previous study (21). Correlation for FM was slightly improved (r= 0.980) and characterized by a smaller offset (52.4 g instead of 122 g) and steeper slope (0.77 instead of 0.69). Again, the conversion equations gave excellent results. The mean residual error for the determination of total calcium and FM was in the range of 0 ± 2 g and 0 ± 40 g, respectively. The smaller offset and steeper slope found in our study may result from the more elaborate DXA drift correction as well as from the homogenization procedure, which included additional lyophilization and a second homogenization step of the dry matter. The 50-g offset and 77% slope may reflect an intrinsic bias of the software currently used. A further software refinement is desirable to get rid of the residual errors. Nevertheless, the reported precision and accuracy seem to be sufficient to justify DXA as a practicable noninvasive in vivo measurement of neonatal body composition in longitudinal and interventional studies.

Assessment of gross energy content.

Gross energy is the total amount of energy stored in the body and reflects the overall nutritional status. The high correlation of gross energy content measured by bomb calorimetry with gross energy content calculated from crude protein and fat suggests that gross energy content is mainly determined by these two body components. The small underestimation of calculated gross energy content may be related to other components (glycogen, DNA, etc.) that contribute to gross energy content but have not separately been determined. Our data indicate that gross energy content may be estimated in vivo with acceptable precision and accuracy using converted DXA data, which is of interest for further nutritional studies in humans or animals.

Small step-phantom.

We were able to show that the baby step-phantom does not introduce a systematic error for the measurement of body composition; it may therefore be used when newborn infants are studied. When the phantom is placed beside the infant, scan width can be reduced. Alternatively, the phantom may be placed below the feet of the baby, which increases scan length but scan width is defined only by the dimensions of the infant. The small tissue bar allows shorter scanning times, which should reduce artifacts caused by child motion.

Hydration of fat-free mass.

Classic data about hydration of fat-free mass were mainly calculated data. In our study, chemical analysis revealed that water content of fat-free mass is not dependent on postnatal age but on BW. We observed a similar finding in adults by measuring body composition using DXA and deuterium dilution (50). This might be explained by the fact that the heterogeneous composition of fat-free mass changes with BW, which is an important finding when lean and fat mass are calculated from the measurement of body water (stable isotope dilution, bioelectric impedance) using a fixed hydration factor [usually set as 0.732 (51)].

Magnetic resonance versus chemical analysis.

MRI uses the fact that fat and muscle tissues have different relaxation times that result in different image intensities, depending on the specific imaging sequence. T1 relaxation in fat is faster than in lean tissue, giving more signal intensity in the T1-weighted sequence.

The correlation of total body fat by MRI and chemical analysis is surprisingly high, because fat layers are also distributed in the intestines and are therefore difficult to assign. The most striking finding, however, is the overestimation of body fat by a factor of 2.5. A simple error in the point-counting procedure, e.g. incorrect calibration of the images, can be excluded inasmuch as body volume correlates well with total BW (r2 = 0.990) and the slope (0.910) represents a reasonable overall specific gravity of the whole body (1.080 g/mL). The number of ambiguous pixels (visceral fat versus lean mass) cannot explain the error because it was far < 1% of the total number of pixels counted. The overestimation of fat is only partially explained by the fact that chemical analysis measures mass whereas MRI assesses volume (specific gravity of fat mass is 0.9 g/mL). The major portion of the overestimation might be caused by partial volume effects, i.e. the fact that fat with relatively strong signal intensity within a voxel gives the impression that this voxel is filled by fat even if its fat content is far < 100%. The problem can be overcome to a certain extent by fine-tuning of the imaging sequence and its parameters.

1H-MRS, on the other hand, uses the fact that fat and water signals have different resonance frequencies and can therefore be separated. To overcome the distortion of fat and water resonances caused by magnetic field inhomogeneities when a whole body is studied, in this study we used 3D-CSI in combination with adequate postprocessing and a spectral fitting technique to asses fat-to-water ratios.

The results obtained by 1H-MRS (3D-CSI) are highly correlated with those obtained by chemical analysis. With an intercept close to zero, MRS slightly underestimates fat-to-water ratios and total body fat when compared with chemical analysis. One possible source of systematic errors might be the value of 0.55 to correct for saturation effects at a relaxation time of 300 ms. Assuming unbiased measurements by chemical analysis and by deuterium dilution, a saturation factor of 0.66 would have resulted in a correct slope of 1.0. Another source for errors could be introduced by differences in the body temperature during acquisition. The study aimed at an identical schedule for all piglets to prevent major differences in the postmortem time before CSI acquisition. Minor differences in body temperature and hence in relaxation times and saturation factors cannot be totally excluded. This effect is not present in neonates having a constant body temperature.

Differences between MRI and MRS are obvious if time required for acquisition and analysis is considered. The preparation of MRI scans is largely done by the MR system after manual prescription of the slice position. The preparation of the 3D-CSI scan is also automatic because all scan parameters are either fixed (e.g. shim values) or taken from the imaging prescan (e.g. radio-frequency flip angle). Time needed for the acquisition of the MRI and MRS data are comparable (15–25 min). The big difference between MRI and CSI, however, shows up during data analysis. Point counting of up to 36 images per piglet is very time-consuming (1.5 h) and introduces operator bias. After a period of program development and setup of fitting constraints, which has to be done only once, postprocessing and fitting algorithms for a CSI data set are almost automatic, without any operator bias, and done in 5–10 min.

Studies using 1H-MRS and chemical analysis have limited the measurement of fat-to-water ratios to smaller animals (rats, chickens, etc.) (26, 35, 36, 39), or animals were scanned slice by slice using either simple slice-selective free induction decay (FID) sequences or 2D-CSI. Correlation coefficients between 0.814 and 0.999 have been found. However, the present study uses 3D-CSI for the first time and makes measurements faster than scanning the animals slice by slice, which requires precise adjustment of homogeneity and radio-frequency angle for each slice.

Conversion of fat-to-water ratios to absolute units in this study has been done by the determination of body water by means of deuterium dilution. However, direct absolute quantitation by MR methods would also be possible provided that signals are calibrated by subsequent acquisition of a calibration probe. If factors such as coil loading, temperature, and so forth are considered, this absolute quantitation could replace additional measurements of body water without major problems.

The 3D-CSI method suggested here faces one inherent limitation. 3D-CSI-data are acquired simultaneously over the whole body and require therefore that the whole body is within the homogeneous volume of the magnet. In our MR system, a diameter of this sphere in the range of 45–50 cm and a comfortable positioning of neonates in such a sphere are possible. A continuous follow-up of older babies, however, may be limited, or they may have to be scanned in multiples of 45 cm.

This study used piglets' bodies, which show neither motion nor blood flow, conditions that may impair data acquisition in neonates. Image quality can be reduced by motion artifacts, but ongoing studies with neonates show that these problems can be solved by sedation, postprandial sleep, fixation, and similar measures. CSI data acquisition is even less prone to motion artifacts.

In conclusion, the data obtained in this study prove that, with some fine tuning (e.g. software refinement), DXA is suited to assess neonatal body composition. The data further indicate that body fat may be estimated without radiation exposure when MR methods are used. Whereas MRI has inherent problems because of appropriate choice of imaging parameters and methods used for volumetry that can hardly be overcome, 3D-CSI MRS seems to be an elegant method and potential candidate to estimate body fat and should be investigated more in detail.