phic model, the later software corrected some anomalies of regional bone mineral content (BMC) observed earlier. There was some slight dependence of total ...
1994, The British Journal of Radiology, 67, 1210-1217
Precision and accuracy of measurements of whole-body bone mineral: comparisons between Hologic, Lunar and Norland dual-energy X-ray absorptiometers 1
P TOTHILL, PhD, FRSE, * 2 A AVENELL, BSc, MRCP and 3 D M REID, MD, FRCP
1
Department of Medical Physics and Medical Engineering, University of Edinburgh, Western General Hospital, Edinburgh EH4 2XU, 2The Rowett Research Institute, Aberdeen AB2 9SB and department of Rheumatology, City Hospital, Aberdeen AB9 8AU, UK
Abstract Measurements of whole-body bone mineral made by Hologic, Lunar and Norland dual-energy X-ray absorptiometers have been compared. It was found that in each case the results were changed by new software protocols introduced by the manufacturers during the course of the study. With a moderately anthropomorphic model, the later software corrected some anomalies of regional bone mineral content (BMC) observed earlier. There was some slight dependence of total BMC on thickness and fat proportion and up to 15% difference between instruments. Measurements on volunteers showed good precision, but there were differences between instruments made by different manufacturers. There were high correlations, but the slopes of regression lines suggested differences of calibration of up to 8%; the standard errors of the estimates were 110 to 190 g, with maximum deviations from regression of 17%. There were regional disparities in BMC, particularly in the trunk, which arise (in part at least) from the imposition of a higher bone threshold by Hologic. From the pattern of results it was concluded that different assumptions were made by the manufacturers, particularly concerning the fat distribution model, which preclude the interchangeability of results from different instruments.
Dual-photon absorptiometry is well established for the measurement of bone mineral in the lumbar spine, hip and other regions of the body. The introduction of an X-ray source instead of the previously used radionuclide sources has brought increased speed and improved precision and resolution, making whole-body scanning more practicable. There have been several reports of assessments of the precision or reproducibility of dual-energy X-ray absorptiometry (DXA), but less attention has been paid to accuracy, that is, the ability to obtain the true answer. Although apparatus for DXA has been introduced commercially and developed considerably, some fundamental limitations of the technique remain. The use of photons of two different energies provides the necessary attenuation measurements to determine two body tissue components independently of thickness. The bone Received 20 December 1993 and in revised form 21 February 1994, accepted 7 March 1994. *Present address: Clinical Biochemistry Department, Aberdeen Royal Infirmary, Aberdeen AB9 2ZB, UK. 1210
mineral component is well defined, but soft tissue is heterogeneous; the attenuation characteristics of fat are such that, in comparison to lean tissue, it appears as "negative bone" in the calculations. Attempts to add a third photon energy to solve the third unknown [1] have been thwarted by the nature of the relationships between attenuation, energy and atomic number [2,3]. It is therefore necessary to make assumptions about the distribution of fat. For measurements of the lumbar spine the simple assumption is usually made that the thickness of fat over the vertebrae is the same as that in the adjacent softtissue background. However, it has been demonstrated that the non-uniformity of distribution of adipose tissue in the abdomen leads to small, but variable and unpredictable errors in the determination of bone mineral density (BMD) of the lumbar spine [4]. For whole-body scanning more complex assumptions are necessary. The dual-energy measurements allow the determination of the proportion of fat in non-bone areas (about 60% of the pixels) and these assessments are used to extrapolate or interpolate over bone. Whatever assumptions are made, they cannot be valid for all subjects, so absolute The British Journal of Radiology, December 1994
DXA instruments compared for whole-body bone mineral measurements
accuracy of bone mineral or soft tissue measurement is not possible. It is also difficult to assess accuracy. In the early development of neutron activation analysis for the measurement of total-body calcium, human cadavers were measured and then the bone mineral was determined chemically [5]. Understandably, few such measurements were made. More would be required to establish the validity of DXA, covering a range of adiposity. There have been some comparisons of bone mineral determinations by dual-photon absorptiometry with measurements of total-body calcium by activation analysis, but these have been limited to normal volunteers and have not included osteoporotic or obese subjects [6,7]. Measurements on animals offer a possibility, but are likely to be limited by the differences in shape, size and tissue composition of animals and humans. Insights into absolute accuracy are probably best obtained by measurements of models (or "phantoms") of known composition, although these can never mimic completely the complexity and variability of the human body. This report includes measurements on one such model. Although accuracy is desirable, the comparability of results from different absorptiometers is almost as important, particularly for multicentre clinical trials and for the interchange of reference data bases. There are three major manufacturers of DXA apparatus, and advantage has been taken of the fact that these are all represented in Scotland to make comparisons between them. Methods
The whole-body densitometers we used were: a Hologic QDR-1000/W in Edinburgh; a Lunar DPX in Glasgow; and two Norland XR 26s in Aberdeen, one a Mark II and the other a Mark II HS (high speed). This last is equivalent to the current XR 36 and will be so designated; it introduces dynamic filtration, altering the intensity of the X-ray beam to compensate for differences of tissue thickness. Results from the two Norland instruments differed only in respect of precision assessment, so that, apart from that, results presented are from the XR 36. The Hologic and Norland have only one speed option for whole-body scanning. In general the Lunar was used at medium speed, but a slower speed was used for thicker objects, as recommended. During the course of the investigation new software was introduced by each of the manufacturers. Although the assumptions made about fat distribution are not revealed, they were evidently changed, leading to different results. For the Hologic machine the initial software version was 5.35, replaced by a Beta test version of Enhanced Whole Body V5.51P in April 1992, confirmed in August 1993 as being the definitive issue. Lunar version 3.4 was replaced by 3.6 in February 1993. Norland version 2.2.5 was replaced by 2.4 in March 1993; the new software included changes in the selection of regions of interest. Fortunately, scans acquired using the earlier software could be re-analysed. Results are Vol. 67, No. 804
presented for all the options, as they illustrate the importance of the assumptions used and are relevant to measurements made by others at different times. In our results the software used is called "new" or "old". In an attempt to assess accuracy absolutely and to illuminate results obtained on people, we used a phantom devised by Nord [8], and lent by Norland Inc. This is moderately anthropomorphic, although without arms, and consists of a simplified skeleton of aluminium sheet with sheets of acrylic (rather fat-like) and thin vinyl (very lean) plastic representing soft tissue. The proportions can be varied to alter the thickness and effective fat proportion. The phantom is illustrated in Figure 1. The minimum fat proportion obtained by including all the vinyl sheets is 22%. To obtain a leaner mix, three sheets of thin aluminium, cut to match the size of the plastic at suitably spaced levels, were added. The maximum thickness of plastic is 15 cm; taking the density into account, this corresponds to approximately 18 cm of soft tissue. To simulate more obese subjects, rectangular sheets of hardboard were added to the trunk in thicknesses of 5 and 10 cm. The equivalent fat content of the hardboard was 40% and its density close to 1 g cm 2. In all the experiments the soft tissue thickness and fat content of the head were unaltered. As it had been found previously in spine scanning that BMC apparently varied with the height of the bone above the couch [9], such a possibility was investigated for whole body scanning by measuring the phantom with the aluminium skeleton underneath, in the middle and on the top of the plastic sheets. To assess comparability in vivo, measurements were made within a short space of time on 11 normal volunteers (six women and five men), covering a range of size and adiposity. Precision was assessed by repeated measurements of the phantom in various configurations. In vivo precision was measured by making two measurements for a number of subjects, with repositioning in between; it was not possible to obtain in vivo precision values for the Lunar machine. Permission for the measurements on volunteers was granted by the ethics committees of the hospitals concerned; in any case, the effective dose is less than 10/iSv, equivalent to about 2 days of natural background.
Figure 1. Nord phantom. 1211
P Tothill, A Avenell and D M Reid 1200 1200 1150-
1100-
—•— - •—±— - A—•— - o-
Hologic.new Hologic.old Norland.new Norland.old Lunar, new Lunar.old
1150-
1100-
•ams
1050-
1050-
CO
otal BMC-
100000 ~£
i—Hologic.new - • - Hologic.old ^—Norland.new - A - Norland.old >—Lunar, new - o Lunar.old
950-
O
1000-
950-
I-
900-
900-
850-
850-
800
800
10
15
20
25
10
20
40
50
60
% fat
Phantom thickness - cm (a)
30
(b)
Figure 2. Total BMC of Nord phantom as measured by the three instruments and with old and new software, (a) Plotted against thickness with constant fat proportion (22%); (b) plotted against fat percentage, with nearly constant thickness (15 cm). Results Phantom Figure 2a shows the total BMC plotted against the thickness of the phantom when the proportion of fat was kept constant at 22% up to a thickness of 15 cm, increasing slightly for greater thicknesses. The thickness quoted is that of the central flat portion of the phantom; the edges are tapered, to simulate a person. Each point is the mean of 3-5 readings. Such a number of repeated readings was necessary in order to obtain sufficient statistical significance of the differences observed. For the Hologic the BMC varied little with the thickness or the software version, but values were below the nominal value of 938 g by about 8%. For the Norland machine also, BMC depended little on thickness, but the change of software led to an increase of about 10%, up to a value close to the nominal. BMC was higher for the Lunar machine and increased with thicknesses of 20 cm or more, largely corrected by the new software. Lunar software version 3.6 offers a "standard" or an "extended research" analysis. The former yielded results that were very similar to those from version 3.4, but the latter differed increasingly with body thickness. The "extended research" option was used for all the new software results presented. 1212
In Figure 2b the phantom thickness is kept constant at 15 cm and the fat proportion varied by removing vinyl sheets. The 22% fat points are the same as the 15 cm points in Figure 2a. With old software, Hologic and Lunar exhibited an apparent increase of BMC with increasing fat which was statistically significant, but not numerically very important. The new software led to a reduction of the slope. Fat proportion hardly affected BMC values for Norland; with new software there was a slight reduction of BMC with increasing fat, comparable to that with Hologic. The analysis procedures allow the determination of regional BMC, and the skeleton was divided into head, trunk and legs. The pattern of results for the trunk and legs was not much different from that for the whole body. However, there were differences for the head, which are illustrated in Figure 3. The most striking finding is the increase of head BMC with thickness of the trunk (thickness over the head being kept constant) for Hologic old software. New software corrected this and gave the value closest to the nominal of 376 g. Lunar, as well as Hologic old software, showed some increase of head BMC with the percentage of fat in the rest of the body. The BMC measured in the head using new software was lower for Norland than for The British Journal of Radiology, December 1994
DXA instruments compared for whole-body bone mineral measurements
440
-•—Hologic.new • - Hologic.old -A—Norland.new A - Norland.old -•—Lunar.new o- Lunar.old
-1
1 ->
'
1
'
1
i
1
' —T
'—1
480 -
O _
— • — Hologic.new - • - Hologic.old —A— •Norland.new - A- Norland.old — • — •Lunar.new - O- Lunar.old
460 _
420
440
D /
/ /
rr''
-
/ •
/
/
CO
|
CO
"O
T3 CO CD
D) 420 -
CO
^
y / / •
400
CO D)
380
'o O
sz 400 o O
CD
CD 380
-f
-
/
.
-
1? V «//' •
/
-
0 /
360
•••
••
^^—-
—A
360 340
^^^*"A-—
340 -
15
20
-
^
>A—'
*"*"" A- — Ai
10
*
.
25
i
1
10
20
.
A-
1
1
1
30
40
50
— 1
•
60
% fat
Phantom thickness - cm
(b) (a) Figure 3. BMC of head of phantom, (a) Plotted against thickness over the trunk for constant fat proportion; (b) plotted against fat percentage with nearly constant thickness over the trunk.
Hologic, whereas the reverse was true for the whole body. Comparisons were also made of areal bone mineral density (BMD) expressed in gem" 2 , i.e. the BMC value divided by the area determined as bone by the software. Dependence on soft tissue thickness or fat proportion was not greatly different from that for BMC, but the ranking was changed. The results are presented in Table I, for the complete phantom, i.e. 15 cm thick and 22% fat. Nominal values are also included; the area was derived from geometrical measurement. Only new software is used, as the original protocol for Norland did not permit the calculation of BMD. The Hologic gave an area measurement close to the nominal, but underestimated BMC and therefore BMD. Norland gave close to the nominal BMC, but overestimated area and therefore underestimated BMD by 25%. Lunar results were high for both BMC and area, so that BMD was only 3% low. The repetition of measurements for each configuration allowed an examination of any possible variation of precision with thickness or fat content. No significant Vol. 67, No. 804
dependence was found, so the results were combined to give an overall figure for each machine; they are included in Table I. Measured BMC did not vary significantly with height of the skeleton above the couch for either Norland or Lunar, but for Hologic it varied by 0.7% cm"1. This is probably not sufficient to compromise in vivo scans, but is a factor that should be considered in phantom measurements.
Table I. Bone mineral parameters for complete phantom, using new software BMC (g)
BMD (g cm" 2 )
Area (cm2)
Hologic Norland Lunar
875 935 1010
1.042 0.886 1.095
840 1055 922
Nominal
938
1.123
835
BMC SD
CV
8 11 11
0.9% 1.2% 1.1%
1213
P Tothill, A Avenell and D M Reid
Measurements on volunteers Results of soft tissue measurements will be reported elsewhere, along with assessments by other techniques; fat proportions are included here only insofar as they influence bone mineral results. For Hologic and Norland the mean total BMC was not altered by re-analysis, but more subtle changes were introduced. The ratio of BMC from new software to that from old was found to drop with increasing proportion of fat in soft tissue. Linear regressions were highly significant, with R = -0.910, p =0.0001 for Hologic, R = -0.866, p= 0.00057 for Norland. For Lunar, new/old ratio did not vary with the proportion of fat, but there was an average drop of 3%. Using new software, the BMC as measured by each machine was compared by linear regression. The parameters for 11 subjects were: Lunar (L) against Hologic (H): L = 0.994H + 200, standard error of the R= 0.974, p=4x\Q-7, estimate (SEE) = 117; Norland (N) against Hologic: N = 0.990H + 370, R = 0.978, p = 1.9x 10"7, SEE = 107; Norland against Lunar: N = 0.922L + 365, R = 0.930, p=3x 10"5, SEE =190. The regressions are highly significant, but the standard errors of the estimates (SEE) are appreciable. The maximum deviations from the regression lines above were, respectively, 277 g, 155 g and 440 g. There were anomalies of regional BMC measurements. Figure 4 presents mean BMC values for arms, legs, trunk and head as a proportion of the total body BMC, using new software. There were some differences of proportions between the sexes: for example, women had a higher proportion of bone mineral in the head than men. More disturbingly, there were differences between machines for legs and trunk. Ratios decreased in the order Hologic, Lunar and Norland for legs, but increased in the same order for trunk. The differences are statistically significant (p < 0.05). To examine whether the differences related to variations in the defined regions of interest, re-analysis of Norland and Hologic scans was undertaken to match visually the equivalent scan on the other machine. This process led to no appreciable change. Altering Hologic scans to give the same leg/total and trunk/total BMC ratios as Norland included too much leg in the trunk region of interest and the converse led to exclusion of part of the pelvis. The use of old software increased the anomalies. To investigate the possibility that ribs and parts of the pelvis might be below the threshold for inclusion in the analysis software as bone, regions of interest were selected manually in the thorax. The results for Lunar and Norland were very similar. Typically, over the ribs, about half of the chosen area would be regarded as bone and a BMD of approximately 0.5 gem" 2 obtained. The BMD from Hologic was about half this, but there was a dramatically lower assessment of bone area and BMC, by a factor of more than 10. Similar results were 1214
0.4-
Figure 4. BMC of regions as proportion of total BMC, using new software.
observed with the Nord phantom. A region of interest of approximately 100 cm2 over the ribs led to a BMD of 0.45 gem 2 and BMC of 15 g with the Norland, but 0.25gem" 2 and 1 g with Hologic. Total body BMD values were compared in the same way as BMC, by linear regression. The parameters observed were: Lunar against R = 0.936, p=2x Norland against R = 0.927, p = 4 x Norland against R = 0.950, p = 8 x
Hologic: L = 0.913H+ 0.143, 10 5, SEE = 0.036; Hologic: N = 1.181H-0.338, 10"5, SEE = 0.050; Lunar: N = 1.240L - 0.459, 10"6, SEE = 0.042.
BMD values for Norland were lower than those for Hologic, although the reverse was true for BMC, because area assessments were, on average, 27% higher for Norland. The use of BMD as a means of normalization for body size may not be ideal in whole-body measurements. An alternative is to use some measure of body habitus. For example, a regression of BMC (B) against weight (W) in kilograms for new software gave: For Hologic: B = 35.7 W + 69, R = 0.93, p = 0.00003, SEE =184; For Lunar: B = 3l.6W + 529, R = 0.81, p = 0.0023, SEE = 303; For Norland: 5 = 3 3 . 8 ^ + 539, R = 0.88, p = 0.00039, SEE = 248. Similar correlations of BMC with height were observed, with R = 0.86 for Hologic, 0.85 for Lunar and 0.86 for Norland. Good linear relationships are observed, but other factors, larger numbers and separation by gender would need to be considered in order to optimize the normalization necessary for setting up a reference data base. An example of the limitations of BMD as a means of normalizing for size is the fact that a regression of BMD against weight is significant, with R = 0.76 for Hologic, 0.81 for Lunar and 0.81 for Norland. The British Journal of Radiology, December 1994
DXA instruments compared for whole-body bone mineral measurements Table II. Reproducibility of repeated scans in people (Hologic, n = 18; Norland XR26 Mkll, n = 7; Norland XR36, n = 9) BMD
BMC Mean (g)
SD (g)
CV (%)
Mean (g c m 2 )
SD (g cm" 2 )
(3V ( 0//o )\
Total body Hologic Norland Mkll Norland XR36
2182 2509 2753
14 34 26
0.7 1.4 1.0
1.089 1.009 0.985
0.007 0.015 0.011
(
Trunk Hologic Norland Mkll Norland XR36
567 858 947
9 29 18
1.6 3.4 1.9
0.839 0.933 0.927
0.009 0.018 0.018
.1 .9 .9
Legs Hologic Norland Mkll Norland XR36
883 867 921
9 13 8
1.0 1.5 0.8
1.209 0.972 0.999
0.010 0.021 0.010
( :1.2 1.0
Arms Hologic Norland Mkll Norland XR36
289 330 370
6 11 7
2.1 3.2 1.8
0.732 0.745 0.662
0.009 0.031 0.020
1.3 U 3.0
In vivo precision was calculated as the mean difference of two measurements on a number of subjects divided by y/2 to give a standard deviation (SD). A coefficient of variation (CV) was SD divided by the mean value, expressed as a percentage. The software version used had little effect on SD; results are presented for new software in Table II. The ability to derive regional BMC is a useful facility in whole-body DXA, so precision was determined for the trunk, legs and arms, as well as total body. It should be noted that different populations were used for the three sets of measurements. Fewer subjects were studied with the Norland machine than with the Hologic and the range of BMC values was smaller. Discussion
The results of precision measurements presented in Tables I and II are generally satisfactory. There are some differences of CV between machines, but larger numbers of repeated measurements, preferably on the same subjects, would be required to establish them firmly. The Norland XR 36 was superior to the XR 26 Mark II densitometer, in spite of its increased scanning speed, because the dynamic filtering maintains an optimum counting rate at the detector. It is to be expected that regional reproducibility would be inferior to the figures for total body; the counting statistics are poorer and there is the added variable of region-of-interest selection. Although we were unable to assess precision of the Lunar DPX in vivo, figures have been published by others. Svendsen et al [10] made two measurements of total body BMC in six adults and reported a CV of 0.9%, which ranks this machine in the same order as our phantom measurements do. Other assessments of precision include that of Herd et al [11], who obtained a CV of 0.6% for total body BMD for a Hologic QDR1000/W, in good agreement with our result. Vol. 67, No. 804
.5 l.l
As far as accuracy is concerned, there are several possible reasons for disagreement between machines in the measurement of bone mineral by DXA. There may be differences of calibration, the dependency of measurement on the thickness of the subject may vary and there are likely to be different assumptions concerning fat distribution. Some or all of these factors may be incorporated in the software used to control the acquisition of data and the subsequent analysis. Manufacturers tend to regard their assumptions and procedures as proprietary, so some of the comparisons reported here may not be readily amenable to explanation. One explicable finding is the disparity in regional BMC values. It has been deduced that Hologic impose a threshold of 0.4 g cm"2 as the lower limit of bone inclusion [12], whereas Lunar and Norland have different approaches to edge detection which include lower levels of BMD. It is difficult to establish the proportion of the trunk skeleton that is not recorded by Hologic, but it evidently contributes to the differences between the instruments. The three manufacturers have different approaches to calibration. Hologic, which derives the two photon energies by rapid switching of the X-ray tube potential, employs a synchronized rotating filter, which includes a sector of calcium hydroxyapatite, to provide internal calibration on a pixel-by-pixel basis. Lunar and Norland both use a constant potential X-ray source and absorption edge filtration to split the spectrum into two parts with different effective energies. They rely on a very stable X-ray source and daily checks with an external standard. Their calibrations differ in that Norland expresses results in terms of calcium hydroxyapatite, whereas Lunar uses "bone mineral", with some allowance for the fact that normal levels of fat in bone marrow falsify the measurement. It is not surprising, therefore, that the measurements of the Nord phantom 1215
P Tothill, A Avenell and D M Reid
presented in Table I and Figures 2 and 3 show mean differences. It is also perhaps not surprising that, at least when new software is used, Norland comes closest to matching the nominal BMC. If the relationships were linear and proportional, differences of calibration could be accommodated by simple correction factors, but our results show that this proviso is not met. For example, total BMC of the phantom measured by the Norland apparatus is intermediate between the values obtained from the other two machines, whereas Norland gives the highest reading in people. There are reasons why measured BMC might be expected to vary with the thickness of soft tissue. The X-ray beams are not monoenergetic, so that increasing thicknesses of absorber will lead to differential attenuation of lower energies and changes in the spectra, which in turn lead to altered assessment of bone mineral. The transmitted photon intensity provides information about thickness traversed, so that correction for beam hardening is possible. Another factor for the systems which use pulse counting (Norland and Lunar) is a possible counting rate limitation. Again, correction is possible. The success of these corrections is illustrated in the phantom results in Figure 2a. There are statistically significant variations of BMC with thickness, but they are relatively minor compared with other uncertainties of measurement. Blake et al [13] studied the effects of beam hardening on bone density measurements with a Hologic QDR1000. Using aluminium sheets in a water tank, they found that BMC varied with water thickness, but that the variation itself depended on the true value of BMD. At BMD values up to 1 g e m 2 , BMD apparently decreased with increasing thickness, whereas the slope reversed at higher BMDs. We found very similar results when studying the effect of soft tissue thickness on spine bone mineral measurements [9]. The slope for Hologic new software in Figure 2 is compatible with these findings. Mazess et al [14] have shown that earlier software used with the Lunar DPX led to some degree of thickness dependence of BMC, but that this was largely corrected by the later version 3.6. They indicate that statistical noise contributed to the apparent increase of BMC at greater thicknesses and that digital filtration in version 3.6 minimized the effect. Our results are in keeping with this report; slower scanning, leading to improved statistics, reduced the excess BMC at thicknesses over 20 cm and the new software gave a further improvement. Also using a Lunar DPX and the earlier software version 3.1 and a simple aluminium sheet simulation of bone, Laskey et al [15] found that BMC and BMD appeared to increase with soft tissue thickness. The possible effects of fat proportion on bone mineral measurements are complex. The effective fat proportion is uniform throughout the Nord phantom. Nevertheless, some dependence of BMC on percentage of fat is evident with the old software for each of the densitometers, as seen in Figure 2b. These errors were largely corrected by new software. 1216
Regional anomalies, as illustrated in Figure 3, were also mostly removed by new software. The head presents a particular problem in DXA. There is not enough soft tissue surrounding the skull to assess the proportion of fat. In any case, it would not be reasonable to extrapolate that value to the brain. It is therefore necessary to assume an average value for the proportion of brain fat in order to derive BMC from the attenuation measurements. It is interesting that, even with new software, Lunar shows some dependence of head BMC on the fat content in the rest of the phantom. Although the use of BMD provides a degree of normalization for body size, there are difficulties in its application to whole body measurements, as there is much overlap of bones in the trunk. Table I demonstrates that the three machines produce different assessments of bone area for the phantom, differences which persisted in measurements on human volunteers. It is somewhat surprising that the Norland machine overestimates the area substantially. The phantom measurements cannot shed any light on the problem of non-uniform fat distribution, which probably leads to the main differences (apart from those of calibration) in bone mineral measurements in vivo by different machines. While noting the disagreements illustrated in the regression equations, we cannot say which results are the most accurate. We can, however, examine how far machines of different manufacture are interchangeable. Simple factors could be used to bring the average whole-body BMC results into agreement, but there would be residual differences, as illustrated by the SEEs and the maximum deviations of the regressions. It seems certain that different machines could not be used in longitudinal studies of individual patients. The results presented here indicate the likely limitations involved in mixing manufacturers for multicentre clinical trials using total body BMC or BMD. The regional BMC results presented in Figure 4 are probably the result of different fat distribution models and the definition of bone-containing pixels; they indicate that in comparing regional measurements even more caution is needed than in whole-body measurements. The study has revealed the importance of the assumptions made and incorporated into the software; we have probably experienced the results of six different sets of assumptions! Manufacturers and their customers face a dilemma regarding improvements in apparatus and methods of data analysis. There is naturally a desire to introduce and use the best options, but a reluctance to make changes frequently. The changes of software reported here produce results sufficiently different to require re-analysis of all earlier whole-body bone mineral measurements to maintain compatibility. Some centres choose to continue using old software for longitudinal studies. There may well be further changes to come; it is vital that manufacturers inform their customers of these changes and their implications. It is also important to bear in mind the fact that assumptions and approximations are unavoidable, and that a universal solution may well be impossible. It would be very desirable to The British Journal of Radiology, December 1994
DXA instruments compared for whole-body bone mineral measurements
6. MAZESS, R B, PEPPLER, W W, CHESNUT, C H ET AL, Total body bone mineral and lean body mass by dual photon absorptiometry. II, Comparison with total body calcium by neutron activation analysis, Calcif. Tissue Int., 33, 361-363 (1981). 7. HEYMSFIELD, S B, WANG, J, HESHKA, S ET AL, Acknowledgments Dual-photon absorptiometry: Comparison of bone mineral We are grateful to Dr Iain Boyle for access to the and soft tissue measurements in vivo with established Lunar apparatus at the Royal Infirmary, Glasgow and methods, Am. J. Clin. Nutr., 49, 1283-1289 (1989). to Mrs Kay Fenner and Dr Rodney Bessent for advice 8. NORD, R H, Soft tissue composition phantom for DXA, Osteoporosis Int., 1, 203 (1991). and assistance in its operation. We should also like to 9. TOTHILL, P and AVENELL, A, Errors in dual-energy thank the volunteers who toured Scotland to be X-ray absorptiometry of the lumbar spine owing to fat measured and Norland Inc (particularly Dr Russ Nord) distribution and soft tissue thickness during weight change, for the loan of the phantom. The work was supported Br. J. Radiol., 67, 71-75 (1994). by a grant to PT from the Advisory Panel on Evaluation 10. SVENDSEN, O L, HAARBO, J, HASSAGER, C and of Medical and Scientific Equipment and Health Services CHRISTIANSEN, C, Accuracy of measurements of body Supplies funded by the Chief Scientific Organization, composition by dual-energy X-ray absorptiometry in vivo, Scottish Office Home and Health Department. AA Am. J. Clin. Nutr., 57, 605-608 (1993). acknowledges the financial support of the Scottish Office 11. HERD, R J M, BLAKE, G M, PARKER, J C and RYAN, P J, Total body studies in normal British women using dual Agriculture and Fisheries Department. DMR is grateful energy X-ray absorptiometry, Br. J. Radiol., 66, 303-308 to the Arthritis and Rheumatism Council for continued (1993). support. 12. KRISHNAN, S S, HARRISON, J E, STRAUSS, A ETAL, In-vitro accuracy and reproducibility of bone and soft References tissue measurements by DEXA. In Human Body Compo1. JONSON, R, ROOS, B and HANSSON, T, Triple-photon sition, ed. by K J Ellis and J D Eastman (Plenum Press, New energy absorptiometry in the measurement of bone mineral, York), pp. 329-330 (1993). Ada Radiol., 29, 461^64 (1988). 13. BLAKE, G M, McKEENEY, D B, CHHAYA, S C ETAL, 2. FARRELL, T J and WEBBER, C E, Triple photon Dual energy X-ray absorptiometry: the effects of beam absorptiometry cannot correct for fat inhomogeneity in hardening on bone density measurements, Med. Phys., 19, lumbar spine bone mineral measurement, Clin. Phys. Phys459^65 (1992). iol. Meas., 11, 77-84 (1990). 14. MAZESS, R B, BISEK, J and TREMPE, J, Effects of tissue 3. KOTZKI, P O, MARIANO-GOULART, D and ROSSI, thickness on fat content and bone using DEXA, Bone, 13, M, Theoretical and experimental limits of triple photon 280 (1992). absorptiometry in the measurement of bone mineral, Phys. 15. LASKEY, M A, LYTTLE, K D, FLAXMAN, M E and Med. Biol, 36, 429^37 (1991). BARBER, R W, The influence of tissue depth and compo4. TOTHILL, P and PYE, D W, Errors due to non-uniform sition on the performance of the Lunar dual-energy X-ray distribution of fat in dual-X-ray absorptiometry of the absorptiometer whole-body scanning mode, Europ. J. Clin. lumbar spine, Br. J. Radiol., 65, 807-813 (1992). Nutr., 46, 39^5 (1992). 5. NELP, W B, PALMER, H E, MURANO, R ET AL, Measurements of total body calcium (bone mass) in vivo with the use of total body neutron activation analysis, J. Lab. Clin. Med., 76, 151-162 (1970). have a better yardstick of accuracy of total bone mineral and soft tissue measurement. A more realistic and variable phantom might help, but its complexity would have to approach that of the human body.
Vol. 67, No. 804
1217
