JOURNAL OF BONE AND MINERAL RESEARCH Volume 6. Number 8, 1991 Mary Ann Liebert, Inc., Publishers
Calibration of Dual-Energy X-ray Absorptiometry for Bone Density RICHARD B. MAZESS, JOEL A. TREMPE, JOSEPH P. BISEK, JAMES A. HANSON, and DIDIER HANS
ABSTRACT Bone mineral content (BMC, g) using DEXA (Lunar DPX) was measured on known hydroxyapatite samples in a water bath in the presence of uniform and nonuniform coverings of fat-equivalent materials. Selective placement of paraffin over bone had a greater effect than lard in reducing apparent BMC, and polycarbonate plastic had a lesser effect. Measured BMC was 100.1 f 1.1% of actual hydroxyapatite weight when (1) fat over bone was about twice the mass of hydroxyapatite, and (2) the surrounding soft tissue was 15-30% fat. There was a linear relationship between observed and expected BMC, area (cm2),and bone mineral density (BMD, g/cm2) measured on an aluminum phantom using either the Lunar DPX or the Hologic QDR1OOO. The measured area with the two densitometers was identical, but BMC differed. For both an anthropomorphic phantom and human subjects, use of a constant-threshold (0.2 g/cm2) edge-detection algorithm excluded less low-density bone from the transverse processes than the standard DPX edge-detection algorithm. Differences in edge detection could influence the results obtained with phantoms and in vivo and make system intercomparison difficult.
INTRODUCTION UAL-PHOTON ABSORPTIOMETRY (DPA) is utilized in about 1000 clinical centers for measurement of bone density of the spine and femur in metabolic bone disease. The DPA method has been shown to be ( I ) accurate on both bone phantoms and bone (2) accurate on cadavers,'J-s) as well as (3) relatively precise (2070) in vivo.(l Recently, densitometers have been developed that replace the IsJGd source of DPA with an x-ray generator or a constant kVp using either switched where the two energies are derived by a K-edge filter that divides the beam into two major components.(lz1 3 ) In general, the results from either switched-energy or K-edge dual-energy x-ray absorptiometry (DEXA) instruments are well correlated to each other and to DPA results ( r > O.95).(+Iz I' There is a 7-15% lower bone mineral density (BMD, g/cm') for some other commercial densitometers (Hologic QDR-1000 and Norland XR-26) compared to some common DPA and DEXA densitometers (Lunar
D
Lunar Corporation, Madison, Wisconsin.
DP3 and DPX). This is due in part to a lower ( 5 % ) bone mineral content (BMC, g), and in larger part to a 5-10% higher area (cm') measurement with these densitometers compared to the Lunar calibration.'6.R.'o.'~) This report examines the calibration of the Lunar DPX on samples of hydroxyapatite, including the effects of fat over bone, to clarify this difference. We also examined the accuracy of area determination on known objects and the influence of edge-detection algorithms. Linearity of response was examined on an aluminum spine phantom that is supplied with each DPX densitometer.
MATERIALS AND METHODS Measurements were done with DPX densitometers (Lunar Corp., Madison, WI) using standard (version 3.2) spine software.[l2)Measurements were made on samples of powdered calcium hydroxyapatite (HDA; density = 3.0 g/cm') of known weight covered by 15 cm of tissue equivalent (water). The hydroxyapatite samples were placed in
800
MAZESS ET AL.
rectangular plastic bags measuring 72 x 100 mm and weighing 18 g. The measurements were made on three (54 g), four (72 g), and five (90g) bags stacked on top of each other. Different levels of lard (99% hydrogenated fat), paraffin, or polycarbonate plastic were placed directly over the hydroxyapatite to simulate the influence of marrow fat. The densities of these fat-equivalent substances were 0.91, 0.92, and 1.20 g/cm3, respectively. We assumed that lard was most equivalent to marrow fat because its composition and density were closest to those for human fat tissue. Paraffin and polycarbonate have attenuation properties similar to those of fat, and could be useful for doing studies or making phantoms. Measurements were made with these samples in a water bath (Fig. 1) covered by 0.0, 2.5, 5.0, and 7.5 cm fat (lard) to simulate different compositions of surrounding soft tissue (0, 13, 23, and 31% fat by weight). Two to four determinations were made for each data point. The standard deviation (SD) in measuring the hydroxyapatite averaged 0.4 g. The coefficient of variation (CV, Yo), or the standard deviation as a percentage of the mean, ranged from 0.4 to 0.8% on the samples. Measurements were made on an aluminum spine phantom (Fig. 2) that simulated four vertebrae differing in bone mineral content (BMC), area, and BMD. This aluminum phantom is supplied with each Lunar DPX densitometer. The BMC per unit volume of aluminum was determined by measuring a bar of aluminum and the hydroxyapatite samples already mentioned with the DPX. There was a BMC of 1.53 g for every cubic centimeter of the aluminum bar. The volume of each segment of the aluminum phantom was calculated from the physical dimensions and multiplied by 1.53 g/cm3 to give the reported BMC. For the Hologic QDR-1000, with an effective lower energy of 45 versus 40 keV for the DPX, the expected cross-calibration to hydroxyapatite would be 4.4% lower (based on the ratio of attenuation coefficients at the two energies) so the conversion factor would be 1.46 g/cm3. A total of 333 determinations were made on 102 different DPX densitometers per phantom (3.3 cases per densitometer). In addition, seven scans of the aluminum phantom were made on a Hologic QDR-1000 (Waltham, MA) densitometer; a factor of I .46 g/cm3 was used to calculate expected values for the QDR. The distribution of BMD in the L2-4 region of the aluminum phantom ( x = 1.25 f 0.01 g/cmz) was compared to that for the L2-4 sequence in five normal subjects ( x 1.23 + 0.12 g/cmz) and five osteoporotic patients ( x = 0.79 0.05 g/cm'). The distribution of bone was also compared to the L2-4 BMD observed with DPX scans of a commercial anthropomorphic phantom ( x = 1.21 0.01 g/cm*) constructed of hydroxyapatite and methyl methacrylate plastic (Hologic, Waltham, MA). The BMC and area of the aluminum and the anthropomorphic phantom are approximately equivalent on the DPX ( - 61 g and 5 1 cm'), but the latter had transverse processes and the former did not.
*
Fat
I
Water
1
Hydroxyapatite
;at
-~ ~
FIG. 1. Experimental setup for measuring hydroxyapa tite (+ fat equivalent) in a water bath. In some cases a uni form fat covering was placed above the water bath.
1 1 1.1 , I D E VIFW
I l,l,N1
iI,
w
1
.
1 .I
FIG. 2. Diagram of the aluminum spine phantom giving dimensions in centimeters. The expected area (cm'), BMC (g), and BMD (glcm') values are given for each segment and for the L1-4 and L2-4 sequences. The area values in parentheses give the area of the "vertebral" segment alone without intervertebral spaces. Each segment contains a small area and BMC contribution from the intervertebral space above and below.
*
-
RESULTS Hydroxyapa t ire The results have been expressed as the mass of fat-equivalent material (lard, paraffin, or polycarbonate) relative to
CALIBRATION OF DEXA
801
the hydroxyapatite mass (Fig. 3). This was done to simulate the effects of differing contents of marrow fat. There was a decrease in BMC as fat equivalent was added; the decrease differed for lard, paraffin, and polycarbonate. With no fat BMD values were overestimated by 9%. When the fat-equivalent mass was equal to the hydroxyapatite mass, the BMC was overestimated by 2% for paraffin, 3% for lard, and 5% for polycarbonate. For measurements without overlying fat, the correct BMC was indicated when the fat-equivalent mass relative to the hydroxyapatite mass was 1.2 for paraffin, 1.6 for lard, and 2.0 for polycarbonate. The results shifted slightly with a uniform fat cover of 2.5-7.5 cm (Table 1 ) . Each 10% increase in overlying fat increased the apparent BMC by about 1 % . With water alone the apparent BMC was 1-2070 low both for paraffin-hydroxyapatite (1.3: 1) and lard-hydroxyapatite (2:1) combinations; for these combinations, the observed BMC averaged 100.1 1.1% of the actual hydroxyapatite weight (over 54-90 g) between 13 and 31% fat.
Aluminum phantom The average BMC, area, and BMD values measured on the aluminum phantoms with the D P X corresponded closely (within 0.1%) to the expected values (Table 2 ) . The
CV among the different determinations was 1--2%for individual vertebral segments and about 0.6% for the L1-4 sequence. For the L1-4 and L2-4 segments the average number of scan lines was 108 and 87, so shifts of 1 scan line made only a slight difference in BMC or area and did not affect average BMD for the total region. However, significant changes in BMC, area, and BMD for small segments were caused by varying placement of intervertebral demarcations. For example, either 20 or 21 scan lines was needed to encompass L1 depending on phantom placement. With 21 traverses there was a 5% greater area, a 3% greater BMC, and a 2% lower BMD for L1. Scans of the same phantom with the QDR-1000 gave area measurements almost identical to those with the D P X . However, the BMC values, and consequently the BMD values, were about 10% lower (Table 2). About half this difference was associated with the higher effective deV of the QDR compared to the D P X (45 versus 40 keV) as the values as a percentage of expected were only 5% low. The distribution of pixels for the aluminum phantom is shown in Fig. 4 in comparison to the distribution for the anthropomorphic phantom, normal subjects, and osteoporotic patients. The two phantoms and normal subjects had similar distribution patterns, but the osteoporotic patients had a much higher percentage of pixels at low densities. The mean BMD in the aluminum and anthropomorphic
Polycarbonate
108
9og
I
Lard
9Og
Paraffin
72g 54g 90g
106 104 102
1
72g
*
54g
100 ,
98
a
v 96
\*
C
\
94
‘ 0.0
1 0.4
’
1 0.8
’
1 1.2
’
1
1.6
,x ‘
’
7 2.0
’ 2.4
Fat Equivalent/BMC
FIG. 3. The measured BMC as a percentage of the actual hydroxyapatite weight, plotted as a function of fat-equivalent mass divided by the hydroxyapatite weight. There was no uniform overlying “fat” in this experiment (see Table I for the effect of overlying fat). BMC declines least for polycarbonate and more steeply for paraffin compared to lard.
802
MAZESS ET AL. TABLE1. MEASURED BMCa
1.3:1 Paraffin/HDA
2:1 Lard/HDA
Fat cover
54 g HDA
cm
Qo
g
0 2.5 5.0 7.5
0 13 23 31
52.5 53.3 53.7 54.2
QO
97.2 98.7 99.4 100.3
72 g HDA g 70.9 71.3 72.2 73.0
90 g HDA
QO
98.5 99.0 100.3 101.4
90 g HDA
&?
%
g
88.6 89.3 90.8 91.5
98.4 99.2 100.9 101.7
89.6 90.2 90.2 90.1
QO
99.6 100.2 100.2 101.1
aBMC measured in g and as a percentage of actual weight on hydroxyapatite (HDA) bags, under 15 cm water plus 2.5 (13%), 5.0 (23%), and 7.5 cm (31%) lard. Samples of 54, 72, and 90 g HDA were measured with a 2: 1 covering of lard, and a sample of 90 g HDA was measured with a 1.3:l covering of paraffin. Four measurements were made for each point, giving a standard error of about 0.2 g.
TABLE2. BMC, AREA,AND BMD OF AN ALUMINUM PHANTOM SIMULATING FOURLUMBAR VERTEBRAE~
LI Lunar DPX BMC Mean, g
cv, 070 070 Expected
Area Mean, cm’
cv, 070 070 Expected BMD Mean, g/cm’
cv, 070 070 Expected Hologic QDR-1000 BMC Mean, g
cv, 070 070 Expected Area Mean, cm’
cv, 070 070 Expected BMD Mean, g/cm2
cv, 070 070 Expected
L2
L3
L4
LI-4
L2-4
8.8 1.02 97.1
12.8 0.84 100.4
17.3 0.80 101.8
22.1 0.76 100.4
61.O 0.63 100.5
52.3 101.1
9.5 0.72 96.4
11.9 0.61 100.2
14.0 0.72 102.0
15.7 0.63 99.8
51.1 0.43 100.2
41.6 0.46 100.9
0.93 0.97 100.8
1.08 0.90 100.2
1.24 0.89 99.7
1.41 1.16 100.7
1.194 0.64 100.3
1.255 0.67 100.0
8.0 1.48 92.6
11.6 0.89 96.0
15.3 94.3
29.0 1.19 95.2
54.8 0.52 94.4
46.9 0.47 95.0
9.6 1.48 98.6
12.1 1.13 102.5
13.8 0.94 100.2
16.0 0.85 102.1
51.5 0.05 101.o
41.8 0.31 101.6
0.83 1.09 94.4
0.96 0.52 93.4
1.11 0.36 94.1
1.25 1.04 93.5
1.06 0.47 93.2
1.12 0.54 93.8
0.64
0.64
aMeasured in 15 cm water on 102 different DPX scanners ( n = 333) and on one Hologic QDR ( n The average results are expressed relative to the expected values and the CV is given.
=
7).
803
CALIBRATION OF DEXA
a
12 11
A
+
10
Normal Oa1mOpOrOtlO Anthropomorphlo phantom Aluminum phantom
9 8 7 6
5 4
3 2 1
0 0.0
0.4
0.8
1.2
1.6
2.0
BMD (g/cm
b
2.4
2.8
z,
100
DO
. +
w
0.0
0.4
0.8
1.2
1.8
Normal Oataoporotls Anthropomorphlo phantom Alumlnum phantom
2.0
2.4
2.8
BMD ( g / c m 2
FIG. 4. The relative (a) and cumulative (b) distribution of areas (i.e., number of pixels) according to BMD. The anthropomorphic and aluminum phantoms were closer to normal subjects (no significant difference) than to osteoporotic patients (p < 0.05).
phantoms did not differ significantly from that in normal subjects, but all these differed from the average BMD for osteoporotics, which was 35% lower. The BMD distribution on the aluminum phantom better approximated the osteoporotics when the L1 + L2 (BMD = 0.97 g/cm’) or L1-3 (BMD = 1.06 g/cm’) sequences were used. For the anthropomorphic spine the BMD for different vertebral segments were similar, so there was no advantage of using L1-3 versus L2-4.
Edge detection The effect of using a constant BMD threshold for detection of bone-containing pixels was examined at three BMD levels (Table 3). A constant-density threshold of 0.2 g/cm’ is used with the Hologic QDR densitometer. The results were compared to those obtained with the DPX. The DPX uses an algorithm incorporating the first derivative (i.e., the slope) of pixels at the bone edge. The total BMC was
MAZESS ET AL.
804
TABLE 3. yo CHANGE FROM TOTALBMC, AREA,OR BMD VALUECOMPARED WITH q o CHANGE WITH LUNAREDGE DETECTION^ YOArea
Yo BMC Normal Threshold (g/cmz) 0.1
0.2
0.3
Lunar
Osteoporotic
Normal
Yo BMD
Osteopororic
Normal
Osreoporotic
Region
x
SD
x
SD
x
SD
X
SD
X
SD
x
SD
Central Ends Both Central Ends Both Central Ends Both Central Ends Both
-0.2 -0.2 -0.2 -0.8 -0.9 -0.8 -1.9 -2.3 -0.2 -1.8 -2.7 -2.2
0.1 0.2 0.1 0.5 0.3 0.5 0.7 0.5 0.7 2.4 3.3 2.7
-0.4 -0.3 -0.4 -2.6 -2.1 -2.4 -5.9 -5.3 -5.7 -6.6 -8.9 -7.5
0.1 0.2 0.1 0.5 0.8 0.5 1.1 1.6 1.2 2.8 2.6 2.3
-1.5 -2.3 -1.7 -5.0 -6.0 -5.0 -9.0 -10.8 -9.3 -11.8 -11.0 -11.5
1.6 2.4
-2.5 -2.3 -2.4 -10.3 -8.7 -9.7 -18.0 -16.4 -17.4 -15.6 -14.6 -15.3
0.8 0.9 0.6 0.9 1.2 0.8 1.1
+1.4 +2.1 +1.6 +4.4 +5.4 +4.4 +7.8 +9.5 +8.0 +11.4 +9.4 +10.6
1.4 2.2 1.3 2.6 2.2 2.6 2.9 1.4 2.8 2.7 3.5 2.6
+2.2 +2.0 +2.1 +8.6 +7.2 +8.1 + 14.7 + 13.3 + 14.2 + 10.9 +7.5 +9.5
0.6 1.2 0.8 1.1 2.1 1.2 1.1 2.1 1.2 4.2 9.6
1.5
3.0 2.5 3.0 3.3 1.9 3.4 3.5 4.7 3.8
1.5
1.3 4.7 8.1 4.5
5.1
aPixels below a fixed threshold (0.1. 0.2, or 0.3, g/cm’ bone) are excluded; the percentage change with Lunar edge detection is given for comparison. For each threshold the results are calculated from the central vertebral region of L2-4, the superior and inferior ends, and both of these. The results are determined from five normal (mean BMD = 1.232 g/cm’) and five osteoporotic cases (mean BMD = 0.788 g/cm’).
lower by 1% in normal subjects and 2% in osteoporotics using a fixed threshold of 0.2 g/cm2. The corresponding figures for area were - 5 and -9%; BMD was thereby increased by 4 and 8% in the two groups with the fixedthreshold approach. The effect was similar for the central vertebral portions and for the ends of the vertebrae, indicating that low-density portions of both the transverse processes and intervertebral regions were affected. In the same groups the edge-detection approach used on the Lunar DPX eliminated more of the low-density bone. As a consequence, the area was reduced by -11% and -15070, respectively, in normal and osteoporotic subjects and the BMD was about 10% higher than that obtained when all bone pixels were included. In other words, the area measured with the conventional DPX approach was about 5% smaller than with the 0.2 g/cmz threshold. Consequently, the BMD was 6% higher in normal subjects and 2% higher in osteoporotics using conventional DPX edge detection versus the 0.2 g/cm’ threshold.
DISCUSSION BMC measurements with Lunar DPA densitometers have been evaluated on excised bones and in studies on cadavers in sit^.('-^) In these studies there was variance of individual sample values about the regression line, which may be due in part to inexact congruence between the densitometric and chemical tests; however, the average BMC measured by DPA did not deviate systematically by more than 2% from vertebral ash weight. Some DPA and DEXA densitometers provide BMD results lower than
those obtained with the Lunar instruments used in these studies. The present results may clarify part of the systematic difference among these densitometers. Lunar bone densitometers have used the University of Wisconsin calibration, which is based on the ashing of bone samples containing marrow fat. Nuclear Data DPA densitometers apparently used this same calibration and gave identical BMC results.(5’Fat-equivalent “marrow” reduces the apparent attenuation due to hydroxyapatite itself. Theoretical considerations indicate the expected decrease is 0.05 g/cmz for each 1 g/cm’ of fat; this effect is similar for 153Gdand for all DEXA densitometers.(’6)Estimates of a larger “fat effect”‘”) are due to erroneous use of an attenuation coefficient for compact bone, which contains collagen, rather than that for pure hydroxyapatite. The theroretical effect has been confirmed in several including the present experimental studies,~b~’l.l*~lb~lO-lO) study. However, we found that polycarbonate, which has been used in some studies,“’) underestimates the effect, and paraffin (and presumably alcohol, which is even lower in density and attenuation) overestimates the effect seen with fat. Our results document the DPX calibration for BMC and show that hydroxyapatite weight is indicated accurately when (1) the fat content is about 1.5 to 2 times that of the ash, and (2) the surrounding fat concentration is 20-30%. This is relevant because (1) the amount of fat in human vertebra averages twice the mass of ash present,I3) and (2) normal subjects have body fat in surrounding soft tissue. Schlemmer et a1.l’’) report that the abdominal fat averaged 30 f 12% in 148 postmenopausal women. Calibration schemes that d o not consider marrow fat at all systematically overestimate BMC in vivo by about
CALIBRATION OF DEXA
805
10%. There is typically 1.5-3.0 g/cm’ of fat in vertebrae, dressed. Our current results show that both DPX and and the BMD varies from about 0.75-1.50 g/cm2. In QDR measurements of area on an aluminum phantom young subjects the ratio of fat to mineral is close to 1, and were exact. The pattern of results we obtained suggested in older subjects the ratio is about 2. For osteoporotic pa- that the area difference observed in vivo appears due to tients the ratio is closer to 4 because fat increases as min- differences in the treatment of low-density bones, such as eral is lost. A calibration that considers marrow fat cannot transverse processes. If so, the construction of phantoms compensate for this wide individual variation, but by as- for intercalibration may require coinsideration of these suming a reasonable fat-mineral ratio (2: I), noninvasive anatomic features. Our results suggest that phantoms could better simulate results have a lesser systematic offset. Calibrations based on the absence of marrow fat, and the density distribution of actual vertebrae in older osteowith the surrounding tissue predominantly fat equivalent porotic patients. A phantom of blocks of annuli, or even (i.e., plastics), could produce BMC (and BMD) results that the aluminum spine phantom, will not address the causes differ. The 10-15% BMD difference between the Hologic of differences in vivo since real spines have ( I ) diffuse den(or Norland) and Lunar DEXA densitometers appears to sity gradients at the edges, (2) transverse processes, and (3) reflect a BMC calibration difference by about 5 ’ 7 0 . ‘ ~ l o 1 4 ) disk spaces. The L2-4 sequence of the aluminum phantom One study with the QDR-1000 showed that the ash weight simulates the bone distribution of normal suhjects, but not on cadaver vertebrae was underestimated by about 6%,‘221 quite as well as the anthropomorphic phantom. However, which is comparable to the Lunar-Hologic difference. In both phantoms are “denser” than osteoporotic patients. the present study, QDR scans on the aluminum spine The L1 + L2 or LI-3 sequences of the aluminum phanphantom produced BMC values that were about 5 % lower tom approximated the BMD values found in older women than those expected based on calculations from attenua- (-1.0 g/cm2) but did not reach the low values seen in tion coefficients. This systematic offset in BMC may re- osteoporosis. The aluminum phantom obviates questions flect different assumptions regarding marrow fat with the about edge detection and thereby facilitates intersystem two densitometers. calibration. Anthropomorphic phantoms allow edge detecArea calibration is the other factor affecting BMD. The tion influences but cannot be effective for cross-calibration area of actual or simulated spines using the Hologic QDR in vivo unless they include both osteoporotic and normal densitometer is about 5-10% greater than with Lunar den- vertebrae. Differing edge-detectioin algorithms produce sitometers.(O l 4 1 5 ’ Edge-detection algorithms may explain different values for area and BMD, depending on bone why there are area differences using different bone scan- density. ners. Use of the fixed threshold of 0.2 g/cm’ on the anAluminum phantoms, such as those we tested, may be thropomorphic phantom gave a BMC of 61 g and an area suitable for relative calibration of bone densitometers of 56 cm’, which were similar to the values specified by the (over time or among scanners) or even absolute calibramanufacturer. Conventional DPX edge detection gave the tion, provided ( I ) a cross-calibration is done against hyarea at 5 1 cm’, or 10% less than the nominal area of 56 droxyapatite, and (2) a defined covering of fat equivalent cm’. The use of different methods for edge detection ap- is used to simulate the physiologic presence of marrow. pears to cause the area difference between instruments; a Moreover, care is necessary in using the same number of fixed threshold of 0.2 g/cm2 excluded less of the transverse scan transverses to encompass specific “vertebral” regions, processes, so the area was higher and the BMD smaller particularly when measurements are done over restricted than when the Lunar edge routine was used. This effect regions of the phantom. The low cost of aluminum phanwould not be evident on objects without transverse pro- toms, the availability of different density sequences, and cesses. the ability to measure them in a medium (water) that better There was virtually no influence of the edge-detection simulates soft tissue than plastics, may offset the fact that algorithm on the aluminum spine phantom. The areas they are (1) not anthropomorphic, (2) not hydroxyapatite, measured with the QDR-1000 corresponded closely to the and (3) unsuitable for addressing effects due to edge detecactual areas and to those measured with the DPX. Using t ion. fixed thresholds of 0. I , 0.2, and 0.3 g/cm2 on the DPX scans of this phantom minimally affected the resultant areas; the corresponding figures were 53.5, 52.2, and 51.5 REFERENCES cm’. There are no “transverse” processes on this phantom, but the edge-detection procedure removed a few low-den1. Mazess RB, Hanson I A , Sorenwn JA, Bardcn H S 19x8 Accuracy and precision of dual-photon absorptiometry. In: sity pixels (probably those due to partial volume artefacts Dequeker J (ed). Proceedings of the Second International at the bone edges). Workshop on Non-lnvasive Bone Measurements. Lxuven We have previously shown that the area of a cylindrical University Press, Belgium, pp. 157-164. annulus with an area of 40 cm2 was measured to within 0.1 2. Schaadt 0, Bohr H 1981 Bone mineral by dual photon abcm’ using the DPX. Further, the projected area of nine sorptiometry. Accuracy- precision -sites of measurements. vertebral phantoms, each containing three vertebrae, was In: Dequeker J, Johnjton CC (eds.) Non-invayive Bone Mealikewise measured to within 0.5% of the area determined surements. IRL Press, Oxford, pp. 59-72. by digitization of radiographs.‘”) However, the transverse 3. Wahner H W , Dunn W , Mazess RU, Towsley M, Lindsay R , processes were deleted from the regions measured on the Markhardt L, Dempster D 1985 Dual-photon (”’Gd) absorpradiographs, so the issue of low-density bone was not adtiometry of bone. Radiology 156:203-206.
806
MAZESS ET AL.
4. Ericksson S, Isberg B, Lindgren U 1988 Vertebral bone min-
eral measurement using dual photon absorptiometry and computed tomography. Acta Radiol 29:89-94. 5 . Gotfredsen A, Podenphant J , Norgaard H, Nilas L, Nielsen VAH. Christiansen C 1988 Accuracy of lumbar spine bone mineral content by dual photon absorptiometry. J Nucl Med 29:248-254. 6. Wahner HW, Dunn WL, Brown ML, Morin R Riggs BL 1988 Comparison of dual energy x-ray absorptiometry
(DEXA) and dual photon absorptiometry (DPA) for bone mineral measurements of the lumbar spine. Mayo Clin Proc
14. Gundry CR, Miller CW. Ramos E, Moscona A, Stein JA, Mazess RB, Sartoris DJ, Resnick D 1990 Dual-energy radio-
15.
16.
17.
63: 1075- 1084. 7. Pacifici R, Rupich R, Vered I . Fischer KC, Griffin M, Susman N, Avioli LV 1988 Dual energy radiography (DER): A preliminary comparative study. Calcif Tissue Int 43: 189-191.
8. Borders J , Kerr E, Sartoris DJ, Stein JA. Ramos E, Moscona AA, Resnick D 1989 Quantitative dual-energy radiographic absorptiometry of the lumbar spine: In vivo comparison with dual-photon absorptiometry. Radiology 170: 129131. 9. O’Malley M, Kenrick AJ, Sartoris DJ, Hochberg AM, Weisman MH, Ramos E, Zvaifler N, Resnick D 1989 Axial bone density in rheumatoid arthritis: Comparison of dual-energy
projection radiography and dual-photon absorptiometry. Radiology 170:501-505. 10. Kelly TL, Slovik DM, Schoenfeld DA, Neer RM 1988 Quantitative digital radiography versus dual photon absorptiometry of the lumbar spine. J Clin Endocrinol Metab 672339844.
I I . Gluer CC. Steiger P , Selvidge R, Elliesen-Kliefoth K, Hayashi C, Genant HK 1990 Comparative assessment of dualphoton absorptiometry and dual-energy radiography. Radiology 174:223-228. 12. Mazess RB, Collick B, Trempe J , Barden HS, Hanson J A 1989 Performance evaluation of a dual-energy x-ray bone densitometer. Calcif Tissue Int 44:228-232. 13. Sorenson JA, Duke PR, Smith SW 1989 Simulation studies of dual-energy x-ray absorptiometry. Med Phys 16:75-80.
18.
19.
graphic absorptiometry of the lumbar spine: Clinical experience with two different systems. Radiology 174539-541. Kelly TL, Slovik DM, Neer RM 1989 Calibration and standardization of bone mineral densitometers. J Bone Miner Res 4:663-669. Sorenson J A 1990 Effects of non-mineral tissues on measurement of boine mineral content by dual-photon absorptiometry (DPA). Med Phys 17(5):90-912. Farrell T J , Webber C E 1989 The error due to fat inhomogeneity in lumbar spine bone mineral measurements. Clin Phys Physiol Meas 10(1):57-64. Cullum ID, Ell PJ, Ryder J P 1989 X-ray dual-photon absorptiometry: A new method for the measurement of bone density. Br J Radiol 62587-592. Hangartner TN Johnston C C 1990 Influence of fat on bone measurements with dual-energy absorptiometry. Bone Miner
9:7 1-82. 20. Hansen MA, Hassager C , Overgaard K, Marslew U. Riis BJ. Christiansen C 1990 Dual-energy x-ray absorptiometry: A
precise method of measuring bone mineral density in the lumbar spine. J Nucl Med 31:1156-1162. 21. Schlemmer A, Hassager C, Haarbo J , Christiansen C 1990 Direct measurement of abdominal fat by dual photon absorptiometry. Int J Obes 14:603-61 I . 22. H o C P , Kim RW, Schaffler MB, Sartoris DJ 1990 Accuracy of dual-energy radiographic absorptiometry of the lumbar spine: Cadaver study. Radiology 176:171-173.
Address reprint requests to: R . B. Mazess, Ph. D. Lunar Corporation 313 West Beltline Highway Madison, WI 53713 Received for publication April 10, 1990; in revised form December 19, 1990; accepted February 8, 1991.
Calibration of Dual-Energy X-ray Absorptiometry for Bone Density RICHARD B. MAZESS, JOEL A. TREMPE, JOSEPH P. BISEK, JAMES A. HANSON, and DIDIER HANS
ABSTRACT Bone mineral content (BMC, g) using DEXA (Lunar DPX) was measured on known hydroxyapatite samples in a water bath in the presence of uniform and nonuniform coverings of fat-equivalent materials. Selective placement of paraffin over bone had a greater effect than lard in reducing apparent BMC, and polycarbonate plastic had a lesser effect. Measured BMC was 100.1 f 1.1% of actual hydroxyapatite weight when (1) fat over bone was about twice the mass of hydroxyapatite, and (2) the surrounding soft tissue was 15-30% fat. There was a linear relationship between observed and expected BMC, area (cm2),and bone mineral density (BMD, g/cm2) measured on an aluminum phantom using either the Lunar DPX or the Hologic QDR1OOO. The measured area with the two densitometers was identical, but BMC differed. For both an anthropomorphic phantom and human subjects, use of a constant-threshold (0.2 g/cm2) edge-detection algorithm excluded less low-density bone from the transverse processes than the standard DPX edge-detection algorithm. Differences in edge detection could influence the results obtained with phantoms and in vivo and make system intercomparison difficult.
INTRODUCTION UAL-PHOTON ABSORPTIOMETRY (DPA) is utilized in about 1000 clinical centers for measurement of bone density of the spine and femur in metabolic bone disease. The DPA method has been shown to be ( I ) accurate on both bone phantoms and bone (2) accurate on cadavers,'J-s) as well as (3) relatively precise (2070) in vivo.(l Recently, densitometers have been developed that replace the IsJGd source of DPA with an x-ray generator or a constant kVp using either switched where the two energies are derived by a K-edge filter that divides the beam into two major components.(lz1 3 ) In general, the results from either switched-energy or K-edge dual-energy x-ray absorptiometry (DEXA) instruments are well correlated to each other and to DPA results ( r > O.95).(+Iz I' There is a 7-15% lower bone mineral density (BMD, g/cm') for some other commercial densitometers (Hologic QDR-1000 and Norland XR-26) compared to some common DPA and DEXA densitometers (Lunar
D
Lunar Corporation, Madison, Wisconsin.
DP3 and DPX). This is due in part to a lower ( 5 % ) bone mineral content (BMC, g), and in larger part to a 5-10% higher area (cm') measurement with these densitometers compared to the Lunar calibration.'6.R.'o.'~) This report examines the calibration of the Lunar DPX on samples of hydroxyapatite, including the effects of fat over bone, to clarify this difference. We also examined the accuracy of area determination on known objects and the influence of edge-detection algorithms. Linearity of response was examined on an aluminum spine phantom that is supplied with each DPX densitometer.
MATERIALS AND METHODS Measurements were done with DPX densitometers (Lunar Corp., Madison, WI) using standard (version 3.2) spine software.[l2)Measurements were made on samples of powdered calcium hydroxyapatite (HDA; density = 3.0 g/cm') of known weight covered by 15 cm of tissue equivalent (water). The hydroxyapatite samples were placed in
800
MAZESS ET AL.
rectangular plastic bags measuring 72 x 100 mm and weighing 18 g. The measurements were made on three (54 g), four (72 g), and five (90g) bags stacked on top of each other. Different levels of lard (99% hydrogenated fat), paraffin, or polycarbonate plastic were placed directly over the hydroxyapatite to simulate the influence of marrow fat. The densities of these fat-equivalent substances were 0.91, 0.92, and 1.20 g/cm3, respectively. We assumed that lard was most equivalent to marrow fat because its composition and density were closest to those for human fat tissue. Paraffin and polycarbonate have attenuation properties similar to those of fat, and could be useful for doing studies or making phantoms. Measurements were made with these samples in a water bath (Fig. 1) covered by 0.0, 2.5, 5.0, and 7.5 cm fat (lard) to simulate different compositions of surrounding soft tissue (0, 13, 23, and 31% fat by weight). Two to four determinations were made for each data point. The standard deviation (SD) in measuring the hydroxyapatite averaged 0.4 g. The coefficient of variation (CV, Yo), or the standard deviation as a percentage of the mean, ranged from 0.4 to 0.8% on the samples. Measurements were made on an aluminum spine phantom (Fig. 2) that simulated four vertebrae differing in bone mineral content (BMC), area, and BMD. This aluminum phantom is supplied with each Lunar DPX densitometer. The BMC per unit volume of aluminum was determined by measuring a bar of aluminum and the hydroxyapatite samples already mentioned with the DPX. There was a BMC of 1.53 g for every cubic centimeter of the aluminum bar. The volume of each segment of the aluminum phantom was calculated from the physical dimensions and multiplied by 1.53 g/cm3 to give the reported BMC. For the Hologic QDR-1000, with an effective lower energy of 45 versus 40 keV for the DPX, the expected cross-calibration to hydroxyapatite would be 4.4% lower (based on the ratio of attenuation coefficients at the two energies) so the conversion factor would be 1.46 g/cm3. A total of 333 determinations were made on 102 different DPX densitometers per phantom (3.3 cases per densitometer). In addition, seven scans of the aluminum phantom were made on a Hologic QDR-1000 (Waltham, MA) densitometer; a factor of I .46 g/cm3 was used to calculate expected values for the QDR. The distribution of BMD in the L2-4 region of the aluminum phantom ( x = 1.25 f 0.01 g/cmz) was compared to that for the L2-4 sequence in five normal subjects ( x 1.23 + 0.12 g/cmz) and five osteoporotic patients ( x = 0.79 0.05 g/cm'). The distribution of bone was also compared to the L2-4 BMD observed with DPX scans of a commercial anthropomorphic phantom ( x = 1.21 0.01 g/cm*) constructed of hydroxyapatite and methyl methacrylate plastic (Hologic, Waltham, MA). The BMC and area of the aluminum and the anthropomorphic phantom are approximately equivalent on the DPX ( - 61 g and 5 1 cm'), but the latter had transverse processes and the former did not.
*
Fat
I
Water
1
Hydroxyapatite
;at
-~ ~
FIG. 1. Experimental setup for measuring hydroxyapa tite (+ fat equivalent) in a water bath. In some cases a uni form fat covering was placed above the water bath.
1 1 1.1 , I D E VIFW
I l,l,N1
iI,
w
1
.
1 .I
FIG. 2. Diagram of the aluminum spine phantom giving dimensions in centimeters. The expected area (cm'), BMC (g), and BMD (glcm') values are given for each segment and for the L1-4 and L2-4 sequences. The area values in parentheses give the area of the "vertebral" segment alone without intervertebral spaces. Each segment contains a small area and BMC contribution from the intervertebral space above and below.
*
-
RESULTS Hydroxyapa t ire The results have been expressed as the mass of fat-equivalent material (lard, paraffin, or polycarbonate) relative to
CALIBRATION OF DEXA
801
the hydroxyapatite mass (Fig. 3). This was done to simulate the effects of differing contents of marrow fat. There was a decrease in BMC as fat equivalent was added; the decrease differed for lard, paraffin, and polycarbonate. With no fat BMD values were overestimated by 9%. When the fat-equivalent mass was equal to the hydroxyapatite mass, the BMC was overestimated by 2% for paraffin, 3% for lard, and 5% for polycarbonate. For measurements without overlying fat, the correct BMC was indicated when the fat-equivalent mass relative to the hydroxyapatite mass was 1.2 for paraffin, 1.6 for lard, and 2.0 for polycarbonate. The results shifted slightly with a uniform fat cover of 2.5-7.5 cm (Table 1 ) . Each 10% increase in overlying fat increased the apparent BMC by about 1 % . With water alone the apparent BMC was 1-2070 low both for paraffin-hydroxyapatite (1.3: 1) and lard-hydroxyapatite (2:1) combinations; for these combinations, the observed BMC averaged 100.1 1.1% of the actual hydroxyapatite weight (over 54-90 g) between 13 and 31% fat.
Aluminum phantom The average BMC, area, and BMD values measured on the aluminum phantoms with the D P X corresponded closely (within 0.1%) to the expected values (Table 2 ) . The
CV among the different determinations was 1--2%for individual vertebral segments and about 0.6% for the L1-4 sequence. For the L1-4 and L2-4 segments the average number of scan lines was 108 and 87, so shifts of 1 scan line made only a slight difference in BMC or area and did not affect average BMD for the total region. However, significant changes in BMC, area, and BMD for small segments were caused by varying placement of intervertebral demarcations. For example, either 20 or 21 scan lines was needed to encompass L1 depending on phantom placement. With 21 traverses there was a 5% greater area, a 3% greater BMC, and a 2% lower BMD for L1. Scans of the same phantom with the QDR-1000 gave area measurements almost identical to those with the D P X . However, the BMC values, and consequently the BMD values, were about 10% lower (Table 2). About half this difference was associated with the higher effective deV of the QDR compared to the D P X (45 versus 40 keV) as the values as a percentage of expected were only 5% low. The distribution of pixels for the aluminum phantom is shown in Fig. 4 in comparison to the distribution for the anthropomorphic phantom, normal subjects, and osteoporotic patients. The two phantoms and normal subjects had similar distribution patterns, but the osteoporotic patients had a much higher percentage of pixels at low densities. The mean BMD in the aluminum and anthropomorphic
Polycarbonate
108
9og
I
Lard
9Og
Paraffin
72g 54g 90g
106 104 102
1
72g
*
54g
100 ,
98
a
v 96
\*
C
\
94
‘ 0.0
1 0.4
’
1 0.8
’
1 1.2
’
1
1.6
,x ‘
’
7 2.0
’ 2.4
Fat Equivalent/BMC
FIG. 3. The measured BMC as a percentage of the actual hydroxyapatite weight, plotted as a function of fat-equivalent mass divided by the hydroxyapatite weight. There was no uniform overlying “fat” in this experiment (see Table I for the effect of overlying fat). BMC declines least for polycarbonate and more steeply for paraffin compared to lard.
802
MAZESS ET AL. TABLE1. MEASURED BMCa
1.3:1 Paraffin/HDA
2:1 Lard/HDA
Fat cover
54 g HDA
cm
Qo
g
0 2.5 5.0 7.5
0 13 23 31
52.5 53.3 53.7 54.2
QO
97.2 98.7 99.4 100.3
72 g HDA g 70.9 71.3 72.2 73.0
90 g HDA
QO
98.5 99.0 100.3 101.4
90 g HDA
&?
%
g
88.6 89.3 90.8 91.5
98.4 99.2 100.9 101.7
89.6 90.2 90.2 90.1
QO
99.6 100.2 100.2 101.1
aBMC measured in g and as a percentage of actual weight on hydroxyapatite (HDA) bags, under 15 cm water plus 2.5 (13%), 5.0 (23%), and 7.5 cm (31%) lard. Samples of 54, 72, and 90 g HDA were measured with a 2: 1 covering of lard, and a sample of 90 g HDA was measured with a 1.3:l covering of paraffin. Four measurements were made for each point, giving a standard error of about 0.2 g.
TABLE2. BMC, AREA,AND BMD OF AN ALUMINUM PHANTOM SIMULATING FOURLUMBAR VERTEBRAE~
LI Lunar DPX BMC Mean, g
cv, 070 070 Expected
Area Mean, cm’
cv, 070 070 Expected BMD Mean, g/cm’
cv, 070 070 Expected Hologic QDR-1000 BMC Mean, g
cv, 070 070 Expected Area Mean, cm’
cv, 070 070 Expected BMD Mean, g/cm2
cv, 070 070 Expected
L2
L3
L4
LI-4
L2-4
8.8 1.02 97.1
12.8 0.84 100.4
17.3 0.80 101.8
22.1 0.76 100.4
61.O 0.63 100.5
52.3 101.1
9.5 0.72 96.4
11.9 0.61 100.2
14.0 0.72 102.0
15.7 0.63 99.8
51.1 0.43 100.2
41.6 0.46 100.9
0.93 0.97 100.8
1.08 0.90 100.2
1.24 0.89 99.7
1.41 1.16 100.7
1.194 0.64 100.3
1.255 0.67 100.0
8.0 1.48 92.6
11.6 0.89 96.0
15.3 94.3
29.0 1.19 95.2
54.8 0.52 94.4
46.9 0.47 95.0
9.6 1.48 98.6
12.1 1.13 102.5
13.8 0.94 100.2
16.0 0.85 102.1
51.5 0.05 101.o
41.8 0.31 101.6
0.83 1.09 94.4
0.96 0.52 93.4
1.11 0.36 94.1
1.25 1.04 93.5
1.06 0.47 93.2
1.12 0.54 93.8
0.64
0.64
aMeasured in 15 cm water on 102 different DPX scanners ( n = 333) and on one Hologic QDR ( n The average results are expressed relative to the expected values and the CV is given.
=
7).
803
CALIBRATION OF DEXA
a
12 11
A
+
10
Normal Oa1mOpOrOtlO Anthropomorphlo phantom Aluminum phantom
9 8 7 6
5 4
3 2 1
0 0.0
0.4
0.8
1.2
1.6
2.0
BMD (g/cm
b
2.4
2.8
z,
100
DO
. +
w
0.0
0.4
0.8
1.2
1.8
Normal Oataoporotls Anthropomorphlo phantom Alumlnum phantom
2.0
2.4
2.8
BMD ( g / c m 2
FIG. 4. The relative (a) and cumulative (b) distribution of areas (i.e., number of pixels) according to BMD. The anthropomorphic and aluminum phantoms were closer to normal subjects (no significant difference) than to osteoporotic patients (p < 0.05).
phantoms did not differ significantly from that in normal subjects, but all these differed from the average BMD for osteoporotics, which was 35% lower. The BMD distribution on the aluminum phantom better approximated the osteoporotics when the L1 + L2 (BMD = 0.97 g/cm’) or L1-3 (BMD = 1.06 g/cm’) sequences were used. For the anthropomorphic spine the BMD for different vertebral segments were similar, so there was no advantage of using L1-3 versus L2-4.
Edge detection The effect of using a constant BMD threshold for detection of bone-containing pixels was examined at three BMD levels (Table 3). A constant-density threshold of 0.2 g/cm’ is used with the Hologic QDR densitometer. The results were compared to those obtained with the DPX. The DPX uses an algorithm incorporating the first derivative (i.e., the slope) of pixels at the bone edge. The total BMC was
MAZESS ET AL.
804
TABLE 3. yo CHANGE FROM TOTALBMC, AREA,OR BMD VALUECOMPARED WITH q o CHANGE WITH LUNAREDGE DETECTION^ YOArea
Yo BMC Normal Threshold (g/cmz) 0.1
0.2
0.3
Lunar
Osteoporotic
Normal
Yo BMD
Osteopororic
Normal
Osreoporotic
Region
x
SD
x
SD
x
SD
X
SD
X
SD
x
SD
Central Ends Both Central Ends Both Central Ends Both Central Ends Both
-0.2 -0.2 -0.2 -0.8 -0.9 -0.8 -1.9 -2.3 -0.2 -1.8 -2.7 -2.2
0.1 0.2 0.1 0.5 0.3 0.5 0.7 0.5 0.7 2.4 3.3 2.7
-0.4 -0.3 -0.4 -2.6 -2.1 -2.4 -5.9 -5.3 -5.7 -6.6 -8.9 -7.5
0.1 0.2 0.1 0.5 0.8 0.5 1.1 1.6 1.2 2.8 2.6 2.3
-1.5 -2.3 -1.7 -5.0 -6.0 -5.0 -9.0 -10.8 -9.3 -11.8 -11.0 -11.5
1.6 2.4
-2.5 -2.3 -2.4 -10.3 -8.7 -9.7 -18.0 -16.4 -17.4 -15.6 -14.6 -15.3
0.8 0.9 0.6 0.9 1.2 0.8 1.1
+1.4 +2.1 +1.6 +4.4 +5.4 +4.4 +7.8 +9.5 +8.0 +11.4 +9.4 +10.6
1.4 2.2 1.3 2.6 2.2 2.6 2.9 1.4 2.8 2.7 3.5 2.6
+2.2 +2.0 +2.1 +8.6 +7.2 +8.1 + 14.7 + 13.3 + 14.2 + 10.9 +7.5 +9.5
0.6 1.2 0.8 1.1 2.1 1.2 1.1 2.1 1.2 4.2 9.6
1.5
3.0 2.5 3.0 3.3 1.9 3.4 3.5 4.7 3.8
1.5
1.3 4.7 8.1 4.5
5.1
aPixels below a fixed threshold (0.1. 0.2, or 0.3, g/cm’ bone) are excluded; the percentage change with Lunar edge detection is given for comparison. For each threshold the results are calculated from the central vertebral region of L2-4, the superior and inferior ends, and both of these. The results are determined from five normal (mean BMD = 1.232 g/cm’) and five osteoporotic cases (mean BMD = 0.788 g/cm’).
lower by 1% in normal subjects and 2% in osteoporotics using a fixed threshold of 0.2 g/cm2. The corresponding figures for area were - 5 and -9%; BMD was thereby increased by 4 and 8% in the two groups with the fixedthreshold approach. The effect was similar for the central vertebral portions and for the ends of the vertebrae, indicating that low-density portions of both the transverse processes and intervertebral regions were affected. In the same groups the edge-detection approach used on the Lunar DPX eliminated more of the low-density bone. As a consequence, the area was reduced by -11% and -15070, respectively, in normal and osteoporotic subjects and the BMD was about 10% higher than that obtained when all bone pixels were included. In other words, the area measured with the conventional DPX approach was about 5% smaller than with the 0.2 g/cmz threshold. Consequently, the BMD was 6% higher in normal subjects and 2% higher in osteoporotics using conventional DPX edge detection versus the 0.2 g/cm’ threshold.
DISCUSSION BMC measurements with Lunar DPA densitometers have been evaluated on excised bones and in studies on cadavers in sit^.('-^) In these studies there was variance of individual sample values about the regression line, which may be due in part to inexact congruence between the densitometric and chemical tests; however, the average BMC measured by DPA did not deviate systematically by more than 2% from vertebral ash weight. Some DPA and DEXA densitometers provide BMD results lower than
those obtained with the Lunar instruments used in these studies. The present results may clarify part of the systematic difference among these densitometers. Lunar bone densitometers have used the University of Wisconsin calibration, which is based on the ashing of bone samples containing marrow fat. Nuclear Data DPA densitometers apparently used this same calibration and gave identical BMC results.(5’Fat-equivalent “marrow” reduces the apparent attenuation due to hydroxyapatite itself. Theoretical considerations indicate the expected decrease is 0.05 g/cmz for each 1 g/cm’ of fat; this effect is similar for 153Gdand for all DEXA densitometers.(’6)Estimates of a larger “fat effect”‘”) are due to erroneous use of an attenuation coefficient for compact bone, which contains collagen, rather than that for pure hydroxyapatite. The theroretical effect has been confirmed in several including the present experimental studies,~b~’l.l*~lb~lO-lO) study. However, we found that polycarbonate, which has been used in some studies,“’) underestimates the effect, and paraffin (and presumably alcohol, which is even lower in density and attenuation) overestimates the effect seen with fat. Our results document the DPX calibration for BMC and show that hydroxyapatite weight is indicated accurately when (1) the fat content is about 1.5 to 2 times that of the ash, and (2) the surrounding fat concentration is 20-30%. This is relevant because (1) the amount of fat in human vertebra averages twice the mass of ash present,I3) and (2) normal subjects have body fat in surrounding soft tissue. Schlemmer et a1.l’’) report that the abdominal fat averaged 30 f 12% in 148 postmenopausal women. Calibration schemes that d o not consider marrow fat at all systematically overestimate BMC in vivo by about
CALIBRATION OF DEXA
805
10%. There is typically 1.5-3.0 g/cm’ of fat in vertebrae, dressed. Our current results show that both DPX and and the BMD varies from about 0.75-1.50 g/cm2. In QDR measurements of area on an aluminum phantom young subjects the ratio of fat to mineral is close to 1, and were exact. The pattern of results we obtained suggested in older subjects the ratio is about 2. For osteoporotic pa- that the area difference observed in vivo appears due to tients the ratio is closer to 4 because fat increases as min- differences in the treatment of low-density bones, such as eral is lost. A calibration that considers marrow fat cannot transverse processes. If so, the construction of phantoms compensate for this wide individual variation, but by as- for intercalibration may require coinsideration of these suming a reasonable fat-mineral ratio (2: I), noninvasive anatomic features. Our results suggest that phantoms could better simulate results have a lesser systematic offset. Calibrations based on the absence of marrow fat, and the density distribution of actual vertebrae in older osteowith the surrounding tissue predominantly fat equivalent porotic patients. A phantom of blocks of annuli, or even (i.e., plastics), could produce BMC (and BMD) results that the aluminum spine phantom, will not address the causes differ. The 10-15% BMD difference between the Hologic of differences in vivo since real spines have ( I ) diffuse den(or Norland) and Lunar DEXA densitometers appears to sity gradients at the edges, (2) transverse processes, and (3) reflect a BMC calibration difference by about 5 ’ 7 0 . ‘ ~ l o 1 4 ) disk spaces. The L2-4 sequence of the aluminum phantom One study with the QDR-1000 showed that the ash weight simulates the bone distribution of normal suhjects, but not on cadaver vertebrae was underestimated by about 6%,‘221 quite as well as the anthropomorphic phantom. However, which is comparable to the Lunar-Hologic difference. In both phantoms are “denser” than osteoporotic patients. the present study, QDR scans on the aluminum spine The L1 + L2 or LI-3 sequences of the aluminum phanphantom produced BMC values that were about 5 % lower tom approximated the BMD values found in older women than those expected based on calculations from attenua- (-1.0 g/cm2) but did not reach the low values seen in tion coefficients. This systematic offset in BMC may re- osteoporosis. The aluminum phantom obviates questions flect different assumptions regarding marrow fat with the about edge detection and thereby facilitates intersystem two densitometers. calibration. Anthropomorphic phantoms allow edge detecArea calibration is the other factor affecting BMD. The tion influences but cannot be effective for cross-calibration area of actual or simulated spines using the Hologic QDR in vivo unless they include both osteoporotic and normal densitometer is about 5-10% greater than with Lunar den- vertebrae. Differing edge-detectioin algorithms produce sitometers.(O l 4 1 5 ’ Edge-detection algorithms may explain different values for area and BMD, depending on bone why there are area differences using different bone scan- density. ners. Use of the fixed threshold of 0.2 g/cm’ on the anAluminum phantoms, such as those we tested, may be thropomorphic phantom gave a BMC of 61 g and an area suitable for relative calibration of bone densitometers of 56 cm’, which were similar to the values specified by the (over time or among scanners) or even absolute calibramanufacturer. Conventional DPX edge detection gave the tion, provided ( I ) a cross-calibration is done against hyarea at 5 1 cm’, or 10% less than the nominal area of 56 droxyapatite, and (2) a defined covering of fat equivalent cm’. The use of different methods for edge detection ap- is used to simulate the physiologic presence of marrow. pears to cause the area difference between instruments; a Moreover, care is necessary in using the same number of fixed threshold of 0.2 g/cm2 excluded less of the transverse scan transverses to encompass specific “vertebral” regions, processes, so the area was higher and the BMD smaller particularly when measurements are done over restricted than when the Lunar edge routine was used. This effect regions of the phantom. The low cost of aluminum phanwould not be evident on objects without transverse pro- toms, the availability of different density sequences, and cesses. the ability to measure them in a medium (water) that better There was virtually no influence of the edge-detection simulates soft tissue than plastics, may offset the fact that algorithm on the aluminum spine phantom. The areas they are (1) not anthropomorphic, (2) not hydroxyapatite, measured with the QDR-1000 corresponded closely to the and (3) unsuitable for addressing effects due to edge detecactual areas and to those measured with the DPX. Using t ion. fixed thresholds of 0. I , 0.2, and 0.3 g/cm2 on the DPX scans of this phantom minimally affected the resultant areas; the corresponding figures were 53.5, 52.2, and 51.5 REFERENCES cm’. There are no “transverse” processes on this phantom, but the edge-detection procedure removed a few low-den1. Mazess RB, Hanson I A , Sorenwn JA, Bardcn H S 19x8 Accuracy and precision of dual-photon absorptiometry. In: sity pixels (probably those due to partial volume artefacts Dequeker J (ed). Proceedings of the Second International at the bone edges). Workshop on Non-lnvasive Bone Measurements. Lxuven We have previously shown that the area of a cylindrical University Press, Belgium, pp. 157-164. annulus with an area of 40 cm2 was measured to within 0.1 2. Schaadt 0, Bohr H 1981 Bone mineral by dual photon abcm’ using the DPX. Further, the projected area of nine sorptiometry. Accuracy- precision -sites of measurements. vertebral phantoms, each containing three vertebrae, was In: Dequeker J, Johnjton CC (eds.) Non-invayive Bone Mealikewise measured to within 0.5% of the area determined surements. IRL Press, Oxford, pp. 59-72. by digitization of radiographs.‘”) However, the transverse 3. Wahner H W , Dunn W , Mazess RU, Towsley M, Lindsay R , processes were deleted from the regions measured on the Markhardt L, Dempster D 1985 Dual-photon (”’Gd) absorpradiographs, so the issue of low-density bone was not adtiometry of bone. Radiology 156:203-206.
806
MAZESS ET AL.
4. Ericksson S, Isberg B, Lindgren U 1988 Vertebral bone min-
eral measurement using dual photon absorptiometry and computed tomography. Acta Radiol 29:89-94. 5 . Gotfredsen A, Podenphant J , Norgaard H, Nilas L, Nielsen VAH. Christiansen C 1988 Accuracy of lumbar spine bone mineral content by dual photon absorptiometry. J Nucl Med 29:248-254. 6. Wahner HW, Dunn WL, Brown ML, Morin R Riggs BL 1988 Comparison of dual energy x-ray absorptiometry
(DEXA) and dual photon absorptiometry (DPA) for bone mineral measurements of the lumbar spine. Mayo Clin Proc
14. Gundry CR, Miller CW. Ramos E, Moscona A, Stein JA, Mazess RB, Sartoris DJ, Resnick D 1990 Dual-energy radio-
15.
16.
17.
63: 1075- 1084. 7. Pacifici R, Rupich R, Vered I . Fischer KC, Griffin M, Susman N, Avioli LV 1988 Dual energy radiography (DER): A preliminary comparative study. Calcif Tissue Int 43: 189-191.
8. Borders J , Kerr E, Sartoris DJ, Stein JA. Ramos E, Moscona AA, Resnick D 1989 Quantitative dual-energy radiographic absorptiometry of the lumbar spine: In vivo comparison with dual-photon absorptiometry. Radiology 170: 129131. 9. O’Malley M, Kenrick AJ, Sartoris DJ, Hochberg AM, Weisman MH, Ramos E, Zvaifler N, Resnick D 1989 Axial bone density in rheumatoid arthritis: Comparison of dual-energy
projection radiography and dual-photon absorptiometry. Radiology 170:501-505. 10. Kelly TL, Slovik DM, Schoenfeld DA, Neer RM 1988 Quantitative digital radiography versus dual photon absorptiometry of the lumbar spine. J Clin Endocrinol Metab 672339844.
I I . Gluer CC. Steiger P , Selvidge R, Elliesen-Kliefoth K, Hayashi C, Genant HK 1990 Comparative assessment of dualphoton absorptiometry and dual-energy radiography. Radiology 174:223-228. 12. Mazess RB, Collick B, Trempe J , Barden HS, Hanson J A 1989 Performance evaluation of a dual-energy x-ray bone densitometer. Calcif Tissue Int 44:228-232. 13. Sorenson JA, Duke PR, Smith SW 1989 Simulation studies of dual-energy x-ray absorptiometry. Med Phys 16:75-80.
18.
19.
graphic absorptiometry of the lumbar spine: Clinical experience with two different systems. Radiology 174539-541. Kelly TL, Slovik DM, Neer RM 1989 Calibration and standardization of bone mineral densitometers. J Bone Miner Res 4:663-669. Sorenson J A 1990 Effects of non-mineral tissues on measurement of boine mineral content by dual-photon absorptiometry (DPA). Med Phys 17(5):90-912. Farrell T J , Webber C E 1989 The error due to fat inhomogeneity in lumbar spine bone mineral measurements. Clin Phys Physiol Meas 10(1):57-64. Cullum ID, Ell PJ, Ryder J P 1989 X-ray dual-photon absorptiometry: A new method for the measurement of bone density. Br J Radiol 62587-592. Hangartner TN Johnston C C 1990 Influence of fat on bone measurements with dual-energy absorptiometry. Bone Miner
9:7 1-82. 20. Hansen MA, Hassager C , Overgaard K, Marslew U. Riis BJ. Christiansen C 1990 Dual-energy x-ray absorptiometry: A
precise method of measuring bone mineral density in the lumbar spine. J Nucl Med 31:1156-1162. 21. Schlemmer A, Hassager C, Haarbo J , Christiansen C 1990 Direct measurement of abdominal fat by dual photon absorptiometry. Int J Obes 14:603-61 I . 22. H o C P , Kim RW, Schaffler MB, Sartoris DJ 1990 Accuracy of dual-energy radiographic absorptiometry of the lumbar spine: Cadaver study. Radiology 176:171-173.
Address reprint requests to: R . B. Mazess, Ph. D. Lunar Corporation 313 West Beltline Highway Madison, WI 53713 Received for publication April 10, 1990; in revised form December 19, 1990; accepted February 8, 1991.
