| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Full paper |
1 Department of Medical Physics, Western General Hospital, Edinburgh EH4 2XU
2 Osteoporosis Research Unit, Woolmanhill Hospital, Aberdeen AB25 1LD, UK
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
Recently, a new generation of DXA instruments has been introduced, using a fan beam geometry to provide faster scanning and improved geometrical resolution. This raises the possibility of errors introduced by magnification effects. We therefore investigated the characteristics of the two main brands using phantoms, or models, to establish their fundamental performance. We also made direct in vivo comparisons for cross-calibration. In addition, comparisons were made with an established pencil beam instrument. As well as magnification effects, calibration of fat and lean tissue and bone mineral measurements were investigated. Factors other than accuracy, such as precision and radiation dose, have not been included as these have been considered elsewhere [11, 12].
A condensed account of some of this work has been published in conference proceedings [13].
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
In vitro measurements
The effects of magnification were investigated by raising phantoms, or their components, to different heights above the couch, with the scanning parameters unchanged. The main phantom used was a thin-walled plastic tank, with slightly tapered sides, 35 cm by 29 cm at its centre, simulating a human trunk. It was filled with water to various depths, from 6–18 cm, the added quantities being measured by weighing. Levitation was simulated by raising the tank on blocks of plastic foam. The effect of the height of bone within the body was examined using an aluminium rectangle, 29 cm by 4.5 cm, simulating a spine, which was suspended in water of 15 cm depth. Similar experiments were performed using a paraffin wax block, 29 cm by 16 cm by 4.5 cm thick, to simulate fat at various heights.
Absolute calibrations were checked against standards recommended by Nord and Payne [14]; that is stearic acid in a block 20 cm by 14 cm by 10 cm thick representing 100% fat, and a solution of 0.6% sodium chloride in water representing 100% lean. The solution was contained in a polyethylene box 21 cm by 15 cm by 8 cm high. The molecular composition of the plastic suggests that in X-ray attenuation terms it should be equivalent to greater than 100% fat. This was tested by scanning a slightly larger water-filled box 20 times with and without the polyethylene box immersed in it. The differences in observed fat proportions, coupled with the measured weights, permitted calculation of the effective fat proportion of the material of the box. The box was also scanned when filled with water alone. The effective composition of the material of the larger rectangular tank was estimated by scanning the partially water-filled tank with and without its lid. As this was composed of the same plastic, the effective composition was calculated from the measured fat values and the weights of the tank, lid and water.
For the 4500 only, a collection of lard in plastic bags, approximately 34 cm wide by 15 cm high by 60 cm long, was scanned with and without a "bone" of aluminium strips, 4.5 cm wide by 50 cm long, at its centre.
Comparisons between the three scanners for soft tissue composition were also made using a variable composition phantom (VCP) similar to that described by Formica et al [15]. This contained five blocks of acrylic plastic (Perspex), each 28 cm by 20 cm by 2 cm, with different degrees of leanness provided by interleaved thin sheets of aluminium instead of vinyl plastic. The total body analysis protocol for the Expert instrument requires there to be a "bone head" in the appropriate position. As its characteristics are unimportant, the Lunar spine phantom was used for this purpose, as suggested by Formica et al [15].
The accuracy of each scanner in measuring changes in bone density was investigated using the whole body phantom described previously [7]. This consists of sheets of tissue-equivalent hardboard cut and stacked as cylinders to simulate a human body, with a simulated skeleton of aluminium strips, the number of which could be changed in the arms, legs and spine to alter the effective bone mineral densities (BMDs).
Measurements were repeated five times. t-tests were used to assess the significance of differences between means, using a cut-off value of p=0.05.
In vivo measurements
Direct comparisons between the absorptiometers were made by scanning a number of subjects on each of two instruments within a short period of time. There was no special selection of the patients or volunteers. A fairly wide range of age, size, adiposity and bone characteristics was achieved. 41 subjects (17 male, 24 female), mean age 32.9 years (SD 14.8 years), mean total body per cent fat 25.3% (SD 11.9%), BMD 1.083 g cm-2 (SD 0.093 g cm-2), were scanned with the 1000 and the 4500; 21 subjects (6 male, 15 female), mean age 42.0 years (SD 13.7 years), mean total body per cent fat 28.8% (SD 9.0%), BMD 1.132 g cm-2 (SD 0.102 g cm-2), were scanned with the 4500 and the Expert. The studies had been approved by the Lothian Research Ethics Committee.
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
Some of the variations of mass with height above the couch were studied at several heights and were found to be linear. For simplicity, the results are summarized in Table 1
for two heights, 0 cm and 15 cm, which cover most of the likely range of heights of body components. Measurements were repeated five times. The significance of height dependence was derived from the differences of mean results at the two heights (p<0.05). Both the 4500 and the Expert showed a dependence on height of the phantom above the couch at approximately 1.3% per cm; this was negative for the 4500 and positive for the Expert, owing to the different positions of the X-ray tubes relative to the couch. The 1000 showed a statistically significant but almost negligible height dependence. The height of the aluminium bone in the tank affected the measured BMC to about the same extent as levitation. With both fan beam instruments, the BMD was affected to a lesser extent, and in the opposite direction, as bone area (BA) changed to a greater extent than BMC. The height of the wax block changed the fat content measured with the 4500, but did not significantly (p=0.06) change it with the Expert.
|
|
|
|
|
|
|
The effective composition of the small polyethylene box was determined to be 105% fat and -5% lean. These nominal proportions relied on DXA scanning. As the mass of the box was small compared with that of its contents, any inaccuracy in the fat measurement would have only a small effect on the estimate of the composition of the combination. When filled with salt solution, the effective fat content was 5% compared with 0% for the solution alone. Results of the calibration experiments are presented in Table 2. Standard deviations on repeated measurements were all less than 1% fat. The effective fat proportion of water alone was taken as 9% [14] and the combination with the box material was calculated as above. Fat equivalence of the acrylic plastic was calculated by RH Nord (Personal communication, 1999). The 1000 underestimates and the 4500 overestimates fat content of the salt solution box. The 4500 appreciably underestimates the fat content of the stearic acid standard, whereas Expert results are intermediate between those from the 1000 and 4500 in all respects and are not far from the nominal values. These relationships were illustrated by calculating regression equations against nominal values. They were highly linear (r>0.999). The equations are included in Table 2. There are appreciable intercepts with both Hologic instruments, but the most notable disparity is the reduced slope for the 4500.
|
as the nominal values. The depth-dependent results for the 4500 agree closely with the nominal values. Values for the 1000 are lower by 7–9%.
The larger phantom of lard measured by the 4500 gave results of 78% fat and 79% fat with and without the simulated bone, respectively. A smaller sample of lard had been compared with the stearic acid block and the fat proportions were not significantly different, so the underestimate of the high fat proportion in Table 2 is confirmed in a phantom closer in size to a human trunk.
The true effective composition of the VCP is not certain. Therefore the observed fat proportions were plotted against the number of aluminium sheets. Regressions over the range 5–65% fat were highly linear (r>0.997) and the equations were:
|
|
|
|
|
|
.
The effective BMD of the aluminium strips used in the hardboard phantom had been previously derived by calibration against calcium hydroxyapatite [7]. When the measured values for each of the variable components of the skeleton were plotted against these nominal values it was found that there was a BMD threshold for the legs of 0.3 g cm-2 with the 4500. A somewhat similar anomaly occurred with each of the bones for the Expert, as both BMC and BA were grossly underestimated at BMDs below 0.4 g cm-2. The upper parts of the plots, above approximately 0.5 g cm-2, were close to linear. Linear regressions were derived; the parameters are presented in Table 3. Also included are the measured values calculated from the regression equations at a nominal BMD of 1.0 g cm-2. Deduced BMDs are all significantly different from unity, and the differences between bones and between instruments are all significant. BMDs and BMCs from the Expert were all higher than those from the 4500. The equations were used to estimate the measured change in BMD or BMC for a 10% increase at an initial BMD of 1.0 g cm-2; these results are included in Table 3. Changes were nearly all underestimated by the 4500. Disparities were less for the Expert.
|
. It is evident that the recorded fat proportion varies with the instrument used, the part of the body considered and the BMD of the bone inserted into the phantom. Apart from legs measured with the Expert, the recorded fat proportion at the lowest BMD was lower than at any higher BMD. Above a BMD of 0.5 g cm-2, the fat proportion recorded by the 1000 was higher than that recorded by the 4500, which was in turn greater than that measured by the Expert, the biggest disparity being 10% fat. With the 4500 and the Expert, there was a small but significant (p<0.05) reduction of fat proportion with increasing BMD. Measurement for the arms differed from the legs for each instrument. The whole phantom had been constructed from the same batch of hardboard sheets, so no differences of composition between the sections were likely.
|
as differences between the means and correlations from linear regression equations. As well as illustrating the differences, the latter serve as cross-calibration formulae, essential in following-up subjects initially measured on the 1000 and subsequently on the 4500. The results were analysed separately for women and men. No significant differences of the regression equations were found, so the combined results for all 41 subjects are presented. The measured mean BMD and BMC are higher with the 1000 compared with the 4500 by up to 11%, and the slopes of the regression equations are greater than unity, with the notable exception of the head. While the means and regression equations should provide sufficient information regarding the differences between the instruments, these are further illustrated by plotting the differences between pairs of results against the mean values for total body BMD, BMC and per cent fat [16] (Figure 2). There are significant biases for BMD and BMC and significant slopes of regression lines for BMC and fat proportion.
|
|
The soft tissue results are presented as fat proportions. Similar conclusions were reached if the comparisons were based on fat masses, but the results were then more subject to differences in ROI selection, especially in the comparisons between the 4500 and the Expert. For the 1000 and 4500 comparisons, there is no mean difference of total fat proportion. This finding results from disparities in opposite directions in different parts of the body by up to 20%, and a balance between slopes that are substantially above unity and negative intercepts in the regression equations. The fat proportion in the head is particularly disparate, with an average difference of 3.5%.
The total body comparisons between the 1000 and the 4500 in vivo are compared with the phantom measurements in Figure 3. The regression lines for the comparisons in vitro are taken from Table 2 and the variable composition phantom Equations (4) and (5) in the text. All regressions had a negative intercept. The slopes are very similar, and all significantly greater than unity.
|
. Again, correlations for males and females were combined as there were no differences between the regression equations. Most of the mean values of BMC and BMD were higher for the Expert than for the 4500 by up to 25%; the head was an exception. In the regressions of Expert against 4500 there were significant positive intercepts for all BMD values and for most BMC values. The slopes were significantly below unity for BMD. Bland–Altman plots [16] for the total body results are presented in Figure 4. There were significant biases for BMD and BMC, and a significant slope for fat proportion.
|
|
. The two comparisons in vitro have similar slopes and intercepts, but the regression in vivo has a steeper slope and a negative intercept.
|
The mean values of fat proportion are not significantly different, except for the arms, but the regression slopes are appreciably greater than unity, with some negative intercepts.
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
, coupled with the relationship between weight and thickness, show that the 4500 fulfills reasonably well the claim by Hologic [17] that, provided the subject remains in contact with the couch, the mass is accurately recorded for all thicknesses. In contrast, the accuracy of mass determined by the Expert varies with water depth. With both fan beam scanners, raising the bath above the couch leads to substantial errors in mass determination (Table 1), although it is unlikely that this would have much influence on normal scanning in vivo. More importantly, Table 1 shows that the height of a bone within the body affects the accuracy of BMC determination, although BMD is little affected because BA changes similarly. These BA changes are, no doubt, associated with procedures to correct for magnification, but there may also be less valid links between BMC and BA, which we have demonstrated previously [18, 19].
In practice, in vivo, the heights of the different components of the skeleton above the couch range from around 2 cm to 20 cm, and with the fan beam scanners the accuracy of the BMC measurements will vary correspondingly. Table 1 suggests that, because of the different X-ray tube orientations, the Expert and the 4500 might differ in, for example, BMC measurements in the arms, where the bones are relatively near the couch. Indeed, Table 5 shows that the Expert/4500 ratio is lower for the arms than for the remainder of the body, in keeping with the above deductions. It is not likely that magnification effects will affect the accuracy of BMC measurements in an individual during weight change, although there are other characteristics of DXA that will do this [18, 19]. However, we have shown that there were significant differences of a few per cent between some bone and fat measurements when a volunteer was scanned prone and supine with the Expert and the 4500 [13].
All of the experiments demonstrate consistent differences in calibration for soft tissue composition between the 1000 and the 4500. The fundamental calibration of DXA instruments stems from the relationship between the measured "r-values", the ratios of the low and high energy attenuations, for known standards such as stearic acid and water. Hologic use such standards [20]. Other materials, such as acrylic plastic and aluminium, are used to monitor calibration during scanning. The same step phantom is used for the 1000 and the 4500. Our phantom measurements provide comparisons of the basic calibrations. The results in Table 2 suggest that in both comparative and absolute terms the 1000 underestimates low fat proportions and the 4500 underestimates high ones. The different slopes are also reflected in the VCP results in Equations (4) and (5), and in the comparisons in vivo in Table 4. They also explain the apparent discrepancies between Table 2, where fat per cent is higher for the 4500 than the 1000 in the water-bath, and Figure 1, where the converse is true for the hardboard phantom. Indeed, the regression equation for total per cent fat in vivo in Table 4 predicts a 1000 measurement of 6% fat compared with a 4500 value of 10%, and 44% fat 1000 compared with 40% fat 4500, values not far from those observed in the phantom measurements.
The final results of scanning in vivo depend on additional assumptions, regarding, for example, fat distribution. The calibration philosophy adopted by Hologic is to adjust QDR 4500A body composition results to provide measurements of per cent fat that are equivalent to per cent fat results generated by a multicomponent model including body water estimates by deuterium dilution, bone mass measurements by DXA and body density measurements by hydrodensitometry [11]. They do not indicate the basis of calibration of the QDR 1000W, but say that per cent fat results from the 4500 are 2–3% lower than earlier QDR scanners at a nominal value of 25%. By direct comparison, we found no difference at this degree of fatness. Kelly [11] gives no indication of whether the difference varies with fat proportion.
The comparisons between a 4500 and a four-compartment hydrodensitometry model have not yet been reported in full. Visser et al [21] report only fat-free mass. There was a high correlation between the two results (r=0.99), but there was a mean fat-free mass 1.8 kg higher by DXA. Total body mass by DXA was 1.1% higher than body weight, so presumably there was good agreement on mean fat mass. No fat proportions are quoted. While errors in lean mass and fat mass are likely to be highly correlated, the data on fat-free mass presented by Visser et al [21] are not incompatible with the discrepancies of fat proportion that we have shown.
The underestimates of low fat proportions by the 1000 in the phantom experiments are supported by our experience that some anorexic subjects have recorded apparently negative per cent fat. Prior et al [22] have compared body composition measurements by a Hologic QDR1000W with a four-compartment hydrodensitometry model in 172 young women and men, using comparable software to ourselves. They found that there was a very good agreement of mean values, but that the 1000 underestimated fat in the leanest women and overestimated it in thefattest, the regression equation being 4C=0.85x1000+3.30, where 4C is the per cent fat from the four-compartment model. This lends support to our conclusions regarding the calibration of the 1000 at low fat proportions. The Prior regression equation for men was less disparate, 4C=0.90x1000+0.75, the intercept being significantly different from that for women.
Arngrimsson et al [23], from the same group, included comparisons between a Hologic QDR 1000W and a four-compartment model in a study of relatively lean subjects of runners and controls. The 1000 consistently underestimated per cent fat, on average by 2.8% fat. All the evidence thus suggests that the 1000 underestimates low fat proportions.
Our conclusion from the phantom measurements that the 4500 underestimates high fat proportions is less well supported. The Prior regression equation predicts that the 1000 overestimates fat proportion in fatter women, but only four subjects had more than 35% fat [22]. From the same centre, Evans et al [2] also compared a four-compartment hydrodensitometry technique with QDR 1000W DXA in 27 obese women. They were primarily interested in measuring changes during dieting and reported only mean results of fat proportions. When all six groups were combined, the mean per cent fat was 43.0% and there was no significant difference between the DXA and four-compartment hydrodensitometry results. On balance, therefore, there is no evidence that the 1000 was inaccurate at high fat proportions in vivo. Coupled with our phantom results and comparisons between the 1000 and 4500 in vivo, we conclude that there may be an underestimate of high fat proportions by the 4500, but that more evidence is required.
Few other direct comparisons of total body measurements in vivo between a QDR 1000W and a QDR 4500 have been reported. Bouyoucef et al [24] included measurements of total body bone mineral on seven subjects, but no regional measurements or soft tissue results. There are disparities between their quoted regression equation and the plot, but it appears that total body BMC and BMD were about 3% higher for the 4500 than the 1000. This is in contrast with our findings that the reverse was true, and emphasizes the fact that such comparisons can only apply to the actual instruments studied. In an abstract from conference proceedings, Fuerst and Genant [25] report comparisons in vivo between a 4500 and a Hologic QDR 2000, used in pencil beam mode. High correlations were found. For BMD, the slope for 4500/2000 was 0.97 (r=0.98). For per cent fat, the slope for 4500/2000 was 0.95 (r=0.99). While the identity between results from the QDR 2000 and QDR 1000W cannot be guaranteed, these results support our findings shown in Table 4 and Figure 4. With an underestimate of 5% fat for the 4500 relative to the 2000, Fuerst and Genant [25] comment that the manufacturer was investigating this apparent incorrect calibration.
There are more uncertainties regarding soft tissue calibration of the Expert. Measurements with the smaller boxes and blocks reported in Table 2 show results intermediate between those from the 4500 and the 1000, with values not far from nominal. However, they do not agree with the composition in vivo (Figure 5).
Results of varying the BMD of the limbs and spine (Table 3) may be compared with those already published for the Hologic QDR 1000 [7]. The 4500 results are similar to those from the 1000. An increase in BMD of 10% is underestimated by both instruments. There is a similar BMD threshold in the legs, which is not so severe with the 4500 as with the 1000. The Lunar Expert gave results closer to the nominal changes than either of the Hologic scanners in measuring BMD changes. BMD measurements with the Lunar Expert were higher than with either Hologic scanner, in keeping with the well known Lunar/Hologic bone calibration differences. The Expert results can also be compared with measurements of the same phantom using a pencil beam Lunar DPX presented in an earlier publication [7]. The main difference is that the Expert bone results were unreliable below a BMD of 0.5 g cm-2, a higher limit than for the Lunar DPX. As a relatively simple bone phantom was used, these results may not be fully representative of the situation in vivo, but they highlight the limitations and differences of the scanners.
A by-product of the variable BMD experiment was the soft tissue composition results presented in Figure 1. Percentage fat differences recorded by the three instruments are in accordance with the calibration experiments, being up to 10% fat between the 1000 and the Expert for the legs. The graphs also demonstrate that the measurement of per cent fat varies with position in the body, the arms showing a higher value than the legs with the Hologic instruments and a lower value with the Expert. The apparent dependence of per cent fat on the BMD, although inaccurate, is understandable at the lowest levels of BMD. Unrecognized bone will contribute to the lean component and lower the recorded value. There is not such an obvious explanation for the smaller but significant reduction of per cent fat with increasing BMD above 0.5 g cm-2 with the Expert and the 4500.
Comparisons of bone and fat measurements in vivo in Tables 4 and 5 demonstrate the need for and value of such cross-calibration. For the two Hologic scanners, the standard errors of 3% for total body BMD and 4% for BMC, while greater than the precision obtainable with either scanner, should allow some longitudinal measurements to span a change of instrument using the cross-calibration equations. Standard errors for the Expert/4500 comparisons are rather higher, but it is less likely that longitudinal studies would involve such a change of brands. The comparisons reported in Table 4 show that the results for the head are markedly different from those for the rest of the body. It has to be remembered that the fat proportion in the brain cannot be measured by DXA as it is surrounded by the skull; thus it has to be assumed and this assumption may also affect bone measurements. The disparities suggest to us that Hologic changed their assumptions in this respect, perhaps as part of a procedure to adjust the total body fat calibration to agree with other techniques.
In conclusion, the phantom measurements have shown that corrections by the manufacturer to deal with magnification effects in fan beam absorptiometers have been reasonably, but not completely, effective. There are calibration differences between the instruments, but the comparisons in vivo provide a means of cross-calibration. Some anomalies remain, which would be worthy of more attention by the manufacturers. It must be remembered that the results apply only to the three particular scanners examined. We have no evidence regarding agreement with other models, nor indeed other examples of the same models, or the effect of different software.
|
Acknowledgments |
|---|
Received for publication March 1, 2000. Revision received July 13, 2000. Accepted for publication September 25, 2000.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  M. J. Bolland, A. B. Grey, G. D. Gamble, and I. R. Reid Low Body Weight Mediates the Relationship between HIV Infection and Low Bone Mineral Density: A Meta-Analysis J. Clin. Endocrinol. Metab., December 1, 2007; 92(12): 4522 - 4528. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M Guler-Yuksel, J Bijsterbosch, Y P M Goekoop-Ruiterman, F C Breedveld, C F Allaart, J K de Vries-Bouwstra, B A C Dijkmans, W F Lems, H K Ronday, A J Peeters, et al. Bone mineral density in patients with recently diagnosed, active rheumatoid arthritis Ann Rheum Dis, November 1, 2007; 66(11): 1508 - 1512. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
W. S. Gozansky, R. E. Van Pelt, C. M. Jankowski, R. S. Schwartz, and W. M. Kohrt Protection of Bone Mass by Estrogens and Raloxifene during Exercise-Induced Weight Loss J. Clin. Endocrinol. Metab., January 1, 2005; 90(1): 52 - 59. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| BJR | DMFR | IMAGING | ALL BIR JOURNALS |
, Males; •, females.
