Influence of Calibration on Densitometric
Studies of Emphysema Progression
Using Computed Tomography
David G. Parr, Berend C. Stoel, Jan Stolk, Peter G. Nightingale, and Robert A. Stockley
Lung Investigation Unit and Wellcome Trust Clinical Research Facility, Queen Elizabeth Hospital, Birmingham, United Kingdom; and Division
of Image Processing, Department of Radiology, and Department of Pulmonology, Leiden University Medical Center, Leiden, The Netherlands
The fundamental importance of calibration for any measuring de-
vice is indisputable, but computed tomography (CT) calibration
in longitudinal lung densitometry studies is largely unexplored.
Although the validity of CT as a measure of emphysema has been
confirmed in cross-sectional studies, there are limited data on long-
term reproducibility, and this is critically important for validating
its use as an outcome measure in therapeutic trials. A general under-
standing of the strengths and pitfalls of CT densitometry is critical
for physicians reviewing the published literature using this method-
ology. In our study of 57 patients with alpha-1 antitrypsin deficiency
(phenotype PiZ), progression of voxel index determined from three
successive annual scans acquired with a fully calibrated scanner was
intimately associated with changes in CT air densitometry, sampled
from patient images. Images were therefore reanalyzed, using a
correction technique validated in phantom studies that adjusted
for changes in measured air density, and the reliability of the voxel
index as a measure of emphysema progression was improved.
Comparison of adjusted voxel index thresholds indicated the opti-
mum threshold was –950 Hounsfield units. Internal air calibration is
therefore critical in longitudinal and multicenter lung densitometry
studies of emphysema and incorporation of a correction factor is
essential for quantitative image analysis.
Keywords: alpha-1 antitrypsin deficiency; emphysema; lung densitometry
Projected estimates of the global burden of chronic obstructive
pulmonary disease have led to recognition of the need for the
development of novel therapies to modify disease progression. In
routine clinical practice, physiologic measures such as diffusion
capacity and airflow obstruction are traditionally employed as
surrogate measures of emphysema severity. However, practical
difficulties arising from the limitations of these conventional
methods (1) necessitate the development of alternative outcome
measures for use in clinical trials (2).
In contrast to chronic obstructive pulmonary disease, which
is defined in physiologic terms, emphysema is defined pathologi-
cally (3). Nevertheless, it may now be diagnosed accurately and
quantified noninvasively by computed tomography (CT). The
loss of lung tissue associated with emphysema can be measured
as a reduction in lung density by CT, using various parame-
ters derived from the frequency distribution histogram of lung
(Received in original form March 11, 2004; accepted in final form July 21, 2004)
ADAPT program supported by a noncommercial grant from Bayer (UK); use of
Pulmo-CMS software supported by European Union funding (grant RNDV.07773).
Correspondence and requests for reprints should be addressed to R. A. Stockley,
M.D., F.R.C.P., D.Sc., Lung Investigation Unit, First Floor, Nuffield House, Queen
Elizabeth Hospital, Birmingham B15 2TH, UK. E-mail: r.a.stockley@bham.ac.uk
This article has an online supplement, which is accessible from this issue’s table
of contents online at www.atsjournals.org
Am J Respir Crit Care Med Vol 170. pp 883–890, 2004
Originally Published in Press as DOI: 10.1164/rccm.200403-326OC on July 21, 2004
Internet address: www.atsjournals.org
voxels. The “voxel index” (VI) method calculates the proportion
of low-attenuation lung voxels within a desired range, and al-
though various voxel index thresholds have been employed and
validated against pathology (4–7), it is likely that no single thresh-
old is appropriate universally.
The fundamental importance of calibration in any measuring
device is indisputable, yet few studies have addressed the issue
of CT calibration in lung densitometry. Short-term reproducibil-
ity has been demonstrated (8–10), but there are limited published
data on long-term reproducibility (11). It is recognized that X-ray
tube ageing and replacement may introduce data acquisition
errors (11), but although accurate scanner calibration is man-
datory, procedures of standardization have yet to be defined.
Previous studies have used water phantoms alone for quality
assurance (4, 8), but in a cross-sectional study to evaluate confor-
mity between scanners, Kemerink and coworkers (12) demon-
strated that although water densitometry was consistent, the
Hounsfield number for air varied between scanners. A method
was proposed for correcting the CT numbers for a scanner with
poor air calibration and its potential use as a means of standardiz-
ing densitometry was described (12). However, although stan-
dardization has been recognized to be crucially important (13),
the influence of air calibration errors on lung densitometry stud-
ies in a clinical setting has yet to be addressed systematically.
Such information is crucial to determine the validity and hence
interpretation of observational and interventional studies.
The provisional assessment of CT progression in a sample of
patients attending our program had unexpectedly suggested an
increase in lung density (consistent with resolution of emphy-
sema) in a limited number of serial annual scans acquired in the
early months of 2000 (L. Dowson, personal communication).
The current study was therefore undertaken to review the consis-
tency of CT densitometry for the measurement of emphysema
progression in a cohort of patients with alpha-1 antitrypsin defi-
ciency (AATD) and to standardize densitometry measurements
by adjusting for any identifiable scanner errors. Validation of the
methodology was performed in studies with an anthropomorphic
phantom. In addition, the opportunity was taken to review data
from the Dutch–Danish controlled trial (14) relating to scanner
calibration.
Some of the results of these studies have been reported pre-
viously in the form of an abstract (15).
METHODS
Data records of 57 patients with AATD (PiZ) were selected from
among those attending our program (see the online supplement) to
include all subjects who had completed three consecutive annual assess-
ments over 2 years spanning the months of concern in early 2000 (see
Introduction). Alpha-1 antitrypsin level and phenotype were verified
by immunoassay and isoelectric focusing, respectively, using a dried
finger-prick blood spot (Heredilab, Salt Lake City, UT). The program
was approved by the local research ethics committee, and all subjects
gave written informed consent.
884 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 170 2004

Annual CT images were acquired on a General Electric Prospeed ity, by Mann–Whitney test. Comparison of the baseline and Year 2
scanner (General Electric Medical Systems, Milwaukee, WI), using a scans at the different thresholds was done with a Wilcoxon signed rank
high-resolution protocol (120 kVp [kilovolt peak], 200 mA · s [dose test for paired data. Differences were considered significant at p ⬍
rate], 1-mm collimation, bone reconstruction algorithm) on inspiration 0.05.
and expiration in the supine position as described previously (16).
The voxel indices (VI) (⫺910 Hounsfield units [HU]) were calculated
for upper (level of the aortic arch) and lower (level of the inferior
RESULTS
pulmonary veins) zone images, using the manufacturer’s Density Mask Baseline Characteristics
program.
The methodology for the current study comprised two steps before The baseline lung function and voxel indices (uncorrected thresh-
the analysis of patient data with an additional phantom study to validate old, ⫺910 HU; expressed as percentage) at baseline are shown in
the technique employed for the patient data. Table 1.
1. Error identification. Consistency of lung densitometry was ex-
plored by plotting annual VI change against scan date. Air mea- Error Identification–Discontinuity Analysis
surements acquired retrospectively from patient images, using A plot of annual VI change against scan date showed a trend
region of interest (ROI) densitometry plus weekly quality assur- occurring in March 2000 of reversed VI progression, implying
ance records of water and bone phantom measurements, were
disease improvement (Figure 1A). The best fitting model for
related to date. Both data sets were assessed for discontinuity to
identify any coincident events. voxel index progression placed this discontinuity (see the online
See Dirksen and coworkers (14), and the online supplement, for supplement) just before March 6, 2000 and produced an esti-
an analysis of the Dutch–Danish controlled trial. mated shift of 4.5% (p ⬍ 0.001). Quality assurance records of
2. Development of a correction method. Air, water, and bone den- scanner calibration showed that measurements using the manu-
sity values were compared to characterize the behavior of the facturer’s water phantom were well within the manufacturer’s
scanner over time across a wide range of density values. A correc- tolerance of 0 ⫾ 10 HU, but a concomitant shift in air density
tion equation was developed to adjust lung densitometry for air measured from patient images was observed (Figure 2A). Re-
calibration errors by predicting the potential influence of these gression analysis of the air calibration data indicated discontinu-
three measures on the VI threshold. ity with a positive shift in air number that occurred between the
Validation Studies Using a Lung Phantom observation on March 6 and the following observation on March
11, 2000 (p ⬍ 0.001) (Figure 2A). In addition, an opposing
Correction technique for longitudinal data. An anthropomorphic lung
drift of decreasing air number also existed throughout the study
phantom was scanned, using the same protocol as in the clinical study,
using a Marconi MxView scanner (Philips Medical Systems, Andover, period, described by the gradient b in the equation y ⫽ a ⫹
MA) with the facility to adjust the CT number for air. Lung densitome- bx1 ⫹ cx2 (see the online supplement). There was no difference
try was calculated for a single slice, using Pulmo-CMS software (MEDIS between the gradient of drift before and after the discontinuity
[Medical Imaging Systems], Leiden, The Netherlands) at different cali- (p ⫽ 0.77) and, after elimination of the discontinuity, the overall
bration settings without the threshold adjustment and applying the gradient drift was significantly different from zero (p ⬍ 0.001)
correction equation to the data to determine their efficacy. (Figure 2B).
Correction technique for cross-sectional multicenter data. Calibra- Measurement of blood density in the descending thoracic
tion and conformity of six third-generation multislice scanners (two
aorta did not identify any time-related trend (data not shown).
General Electric Lightspeed scanners, one Philips MxView, one Philips
MX 8000, one Toshiba Asteion, and one Siemens Sensation 16) were
Dutch–Danish Study
explored, using the dog phantom. The scanning protocol employed
was that proposed for future clinical densitometric studies (140 kVp; A similar drift but no obvious discontinuity was seen by analyzing
40 mA; 5-mm collimation; pitch, 1.5; 2.5-mm increment; field of view, ROI air measurements in the Dutch patients included in the
500 mm; and a smooth reconstruction filter) (17). Air calibration mea- Dutch–Danish study of augmentation therapy (14). The individ-
surements were acquired as described above and surrogate measures ual patient data are shown in Figure 3.
of blood density were derived from ROI measurements of the phantom
heart.
Whole lung VI of ⫺910 HU was calculated for each series with and
without the use of an internal calibration procedure and the effect on
standardization was assessed by the coefficient of variation. TABLE 1. BASELINE CHARACTERISTICS
n Absolute Value* % Predicted*
Analysis of Clinical Images: Employing the Validated
Correction Technique Age, yr 57 50 (46–61) NA
FEV1,† L 57 1.34 (0.97–2.08) 42 (30–67)
Upper-zone inspiratory images obtained previously were reanalyzed, VC,† L 57 3.82 (3.12–4.49) 98 (90–127)
using the “adjusted” threshold of ⫺910 HU. Comparison was made of RV,† L 57 2.66 (2.06–3.44) 135 (100–170)
the annual VI progression rate, using uncorrected and adjusted thresh- DLCO, mmol/min/kPa 57 5.90 (4.62–7.64) 63 (47–84)
olds for three groups of consecutive annual scans: (1 ) scans predating DLCO/VA, L mmol/min/kPa/L 57 1.03 (0.73–1.31) 67 (46–83)
the discontinuity, (2 ) scans spanning the discontinuity, and (3 ) scans UZI voxel index, –910 HU 57 34.2 (21.6–44.6) NA
following the discontinuity. UZE voxel index, –910 HU 57 19.0 (7.7–34.2) NA
In addition, the VI was calculated for corrected images, using 20-HU LZI voxel index, –910 HU 57 55.3 (33.2–64.5) NA
threshold increments from ⫺870 to ⫺950 HU and comparison of pro- LZE voxel index, –910 HU 57 47.0 (26.4–57) NA
gression over 2 years was performed.
Definition of abbreviations: DLCO ⫽ single-breath diffusing capacity of the lung
Statistical Analysis for carbon monoxide (gas transfer); DLCO/VA ⫽ DLCO normalized per liter alveolar
volume (transfer coefficient); HU ⫽ Hounsfield unit; LZE ⫽ lower-zone expiratory
Data were analyzed with a statistical software package (Statistical Pack- high-resolution computed tomography (HRCT) image; LZI ⫽ lower-zone inspira-
age for the Social Sciences [SPSS] version 10.0.5; SPSS, Chicago, IL). tory HRCT image; NA ⫽ not applicable; RV ⫽ residual volume; UZE ⫽ upper-
Air calibration was assessed with discontinuity analysis and regression zone expiratory HRCT image; UZI ⫽ upper-zone inspiratory HRCT image.
slopes were compared with the F test. The rate of CT progression in * Median (interquartile range).
the three subgroups of scan pairs was compared by Kruskal–Wallis test †
Postdual bronchodilation with nebulized salbutamol (5 mg) and ipratropium
and in the two subgroups of scan pairs, before and after the discontinu- bromide (500 ␮g).
Parr, Stoel, Stolk, et al.: Calibration in CT Lung Densitometry
Figure 1. Annual progression rate in upper-zone inspiratory (UZI) scans
at a threshold of ⫺910 HU before (A ) and after (B ) correction for air
calibration. Solid triangles ⫽ voxel index (VI) progression between base-
line and the first year; open circles ⫽ voxel index progression between
the first and second years. Emphysema progression is indicated by points
lying above zero. The arrows indicate the period when the reversal of
the trend (emphysema improvement) was suspected.
Development of a Correction Method
Changes in air, water, and bone density values indicated that
the effect of an increase in air number would be to increase
proportionally the Hounsfield number of those voxels with a
density intermediate between water and air (see Figure 4A).
Therefore, correction for air calibration errors on VI thresh-
old could be achieved with the following formula: corrected
threshold ⫽ (air calibration value ⫺ water calibration value /
⫺1,000) ⫻ index threshold (see Figure 4B).
Validation Studies with Lung Phantom
Correction technique for longitudinal data. The distribution of
voxels with the dog lung phantom was similar to that seen in
patients with AATD (data not shown), suggesting that it was a
good model for emphysematous lung. Adjusting the scanner air
calibration (see Methods) resulted in a proportional shift in the
lung voxel distribution histogram and marked changes to the
voxel index (⫺910 HU) as predicted from the studies described
above. Application of the correction technique reproduced the
same results at each calibration setting (see Table 2).
Correction technique for cross-sectional multicenter data. In-
terscanner variability for air densitometry was greater than for
blood densitometry and there was wide discrepancy in the mea-
surement of dog lung voxel indices (VI, ⫺910 HU). There was
a significant reduction in this discrepancy by standardizing for
885
Figure 2. (A ) CT numbers for air (open circles) and water (solid circles)
over time. CT numbers for water (taken from region of interest [ROI]
measurements of quality assurance water phantom) show consistency
over time. In comparison, the trend in ROI measurements of ambient
air (taken retrospectively from apical images) shows a gradual long-
term negative drift, interrupted by a positive shift occurring in March
2000. Discontinuity analysis of air number data, using the equation
y ⫽ a ⫹ bx1 ⫹ cx2, gives a value for x1 of 6/3/00 to 11/3/00 (p ⬍
0.001), where y is air density value, x1 is scan date (measured as days
since 31/12/1899), x2 is 0 before shift and 1 after the air calibration
shift, a is the y intercept, b is gradient (or “drift”), and c is the shift in
air calibration. Comparison of the gradients of negative drift before and
after the discontinuity by residual sum of squares showed no significant
difference (p ⫽ 0.77). (B ) Plot of ROI measurements for air, adjusted
to eliminate the effect of the shift (c). The gradient (b) is significantly
different from zero (p ⬍ 0.001).
air calibration, using the method described here and used in the
main study, but the use of blood calibration did not reduce
variability (see Table 3).
CT Progression: Application of the Correction Method
The observed reversal in annual voxel index progression that
occurred in the patient cohort was eliminated on use of the
correction technique (see Figure 1B). For the uncorrected data,
the progression rate for scan pairs spanning the discontinuity
was significantly different (p ⬍ 0.001) from that of the other
two groups, in which the scan pair either predated or postdated
the discontinuity (see Figure 5A) and suggests an overall im-
provement over this time period. This difference was abolished
when the corrected data for the three groups were compared
(p ⫽ 0.13) (see Figure 5B). However, there was no significant
difference in VI progression rate for scan pairs obtained either
before or after the discontinuity in both the uncorrected (p ⫽
0.92) and corrected data (p ⫽ 0.72) (see Figures 5A and 5B).
886 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 170 2004
Figure 3. Air calibration data for the Philips Tomoscan SR7000 scanner
used in the Dutch–Danish trial of alpha-1 antitrypsin augmentation
therapy (14). ROI measurements were acquired from patient images
as described for the current study (see METHODS and online supplement).
Air density values (in Hounsfield units) are expressed as the difference
from baseline values and results from individual patients are shown by
the joined lines.
Adjusting the analysis threshold for air calibration also im-
proved the detection of progression over 2 years as shown by a
comparison of Z statistic (see Figure 6). Progression of the voxel
index at the uncorrected threshold of ⫺910 HU was statisti-
TABLE 2. SINGLE IMAGE VOXEL INDEX AT A THRESHOLD
OF ⫺910 HU FOR DOG LUNG PHANTOM
Adjusted Air Calibration
⫺1,000 HU ⫺975 HU ⫺950 HU ⫺925 HU ⫺900 HU
Uncorrected VI 30.27 7.91 1.05 0.08 0.00
Corrected VI NA 30.29 30.43 30.74 30.90
Definition of abbreviations: NA ⫽ not applicable; VI ⫽ voxel index.
Table shows the effects of changing air calibration and the corrected value
achieved through application of the formula presented in text.
cally significant only for the upper-zone inspiratory scans (see
Figure 6A) as shown previously (8). However, after adjustment
with correction for air calibration, significant progression at the
threshold of ⫺910 HU was confirmed in both upper- and lower-
zone scans (see Figure 6B). The annual VI progression rates at
a corrected threshold of ⫺910 HU are shown in Table 4, ex-
pressed as median and interquartile range, with data from the
previous study (8) for comparison.
The upper-zone voxel index showed significant progression
at all thresholds used, with marginal statistical superiority of
the inspiratory images over the expiratory images (Figure 6A).
Furthermore, in the lower-zone scans the corrected expiratory
images now also showed significant progression whatever the
threshold and the corrected inspiratory images demonstrated
significant progression at thresholds of ⫺950, –930, and ⫺910 HU
(Figure 6B). Although all thresholds showed highly signifi-
cant progression of VI, the Z statistic was best for ⫺950 HU,
Figure 4. Plot of actual CT numbers for air, water phantom,
and bone phantom on a single occasion, demonstrating
correct water densitometry but incorrect values for air and
bone. Perfect calibration values are shown for comparison.
(A ) All data points; (B ) the influence on low-density values,
assuming linearity between air and water. The effect of a
shift in air calibration of ⫹30 to ⫺970 HU on an intended VI
threshold of ⫺910 HU is shown by the arrow, demonstrating
how this adjusted threshold can be calculated, assuming a
water value of zero, using the following formula: corrected
threshold ⫽ (air calibration value ⫺ water calibration value/
⫺1,000) ⫻ index threshold.
Parr, Stoel, Stolk, et al.: Calibration in CT Lung Densitometry
TABLE 3. MEAN, STANDARD DEVIATION, AND COEFFICIENT OF VARIATION FOR BLOOD
AND AIR DENSITY VALUES MEASURED ON SIX DIFFERENT SCANNERS
Blood Density (HU) Air Density (HU)
Mean 41.09 ⫺1,005.1
Standard deviation 5.36 13.73
Coefficient of variation 13.04 1.37
Voxel index values (VI, ⫺910 HU) for whole dog lung are shown, demonstrating the influence of standardization on scanner
variability by calibrating on the basis of either blood or air densitometry. The latter resulted in an almost threefold reduction in
variability.
suggesting it is the most appropriate for assessing progression
among the patients studied here.
DISCUSSION
The current study demonstrates the importance of precise scan-
ner and image calibration in longitudinal lung densitometry stud-
ies for the measurement of emphysema progression. We found
Figure 5. Annual progression in upper-zone inspiratory voxel index
(threshold, ⫺910 HU) (UZI VI) before (A ) and after (B ) adjustment for
air calibration. Data are shown as the median and interquartile range.
p Values relate to the Kruskal–Wallis test comparing VI progression rate
in the scan pairs spanning the discontinuity with the other two groups
and to the Mann–Whitney U test comparing scan pairs before and after
the discontinuity. For the unadjusted data (A ) there is a significant
difference in the rate of VI progression between the scan pairs spanning
the time of calibration discontinuity and the scan pairs before and after
the discontinuity (p ⬍ 0.001), but no significant difference between
scan pairs before and scan pairs after the discontinuity (p ⫽ 0.92). (B )
After adjustment for air calibration there is no significant difference
between the scan pairs spanning the discontinuity and the other two
groups (p ⫽ 0.13) or between the groups before and after the disconti-
nuity (p ⫽ 0.72).
887
VI, ⫺910 HU
No Calibration Blood Calibration Air Calibration
36.23 35.56 34.02
8.95 9.27 3.33
24.70 26.06 9.79
that voxel index progression in a cohort of patients with AATD
was associated with changes in air density measurements ob-
tained retrospectively from patient images. A clear shift occurred
in air calibration in March 2000 that produced an obvious rever-
sal in voxel index progression that not only aroused our initial
suspicion of an error, but also allowed us to causally relate the
two events. An isolated drift in air calibration would probably
have remained undetected without specific observation using an
ROI calibration and the effect on VI progression would have
been one of overestimation. No significant difference was found
between the rate of air calibration drift before and after March
2000 and this is reflected in the comparable rates of apparent
VI progression during these periods.
The underlying cause of the changes to our scanner calibra-
tion remains uncertain despite careful inspection of the service
records. Although the most likely cause for our observations is
the replacement of the X-ray tube and detectors, which occurred
in January 2000, no definite explanation can be offered for the
calibration errors. Importantly, the changes in measured air den-
sity occurred despite the apparent stability of CT calibration
when defined by the CT numbers for air and water obtained
during regular routine scanner maintenance and from the weekly
quality assurance measurements using the manufacturer’s water
phantom. A similar finding has previously been reported in a
longitudinal study documenting changes in measurements of
aortic blood density, where full routine scanner calibration was
documented monthly (11). Measurements obtained from patient
images therefore provide essential calibration information in
addition to routine procedures in which densitometry is applied
to simple water phantoms or air measurements obtained with
an empty gantry.
The scanner used in our study was a third-generation single-
slice helical CT. Although technological improvements and the
incorporation of features such as daily autocalibration into mod-
ern scanners may improve the accuracy and precision of lung
densitometry studies, these procedures ensure only a value of
⫺1,000 HU for air with an empty gantry, which may engender
its own problems. For instance, using the technique described
in the current study, we have identified a similar drift in air
calibration in other scanners, including the Philips Tomoscan
SR7000 scanner (see Figure 3) that was used in the Dutch–Danish
trial of alpha-1 antitrypsin augmentation therapy (14). These
errors occurred in spite of the regular and frequent procedure
of CT number adjustment for air calibration that was performed
with an empty gantry. The drift is likely to have influenced the
consistency of lung densitometry because we have demonstrated
a close association between these measures and VI changes. This
provides further evidence of the importance of including an
internal calibration procedure to adjust for potential errors
in air densitometry in longitudinal studies of emphysema. The
error correction described in our study overcame the problem
of drift in air calibration, providing a more accurate measure of
888 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 170 2004
emphysema progression. Our results confirm that our previous
finding of highest VI progression rate in the upper-zone inspira-
tory images (8) remains valid after adjustment of scanner calibra-
tion errors. Furthermore, adjustment for air calibration resulted
in improved variability and the detection of statistically significant
progression in all images, even at the threshold of ⫺910 HU used
in our original study (8). Consequently, this correction method
has now been incorporated in the analysis software being used
for our clinical trials and includes calibration of the entire image
data before lung segmentation and histogram analysis (18).
Adjustment of lung densitometry to correct for errors in air
calibration provides a mechanism to improve scanner consistency
in longitudinal studies and scanner comparability in multicenter
studies as validated in the phantom studies. The voxel distribution
histograms for the dog lung phantom compared well with those
from patients with emphysema, confirming that the phantom
was an appropriate model to validate the correction method.
Although the scanning protocols differed between clinical and
phantom studies, it was necessary in the multicenter phantom
study to use a whole lung volume scanning protocol and whole
lung densitometry to reduce sampling errors.
The multiscanner phantom study confirms the findings of
Kemerink and coworkers (12) that consistency of water calibra-
Figure 6. Voxel index progression over 2 years in up-
per-zone images (A ) and lower-zone images (B ), ex-
pressed as a Z statistic obtained by Wilcoxon signed
rank test. Adjusted thresholds are indicated in Houns-
field units. The threshold for uncorrected data at ⫺910
HU is shown for comparison. Solid boxes represent ex-
piratory images, and open boxes represent inspiratory
images. The significance of the change over 2 years is
shown for all data. Data labels refer to p values.
tion between scanners coexists with great variability in air cali-
bration. It is not surprising, as the density of blood is close to
that of water, that interscanner variability in lung densitometry is
not improved by an internal calibration method that incorporates
blood density values alone whereas variability is improved when
adjusting for differences in air calibration. However, the applica-
tion of blood calibration may be of greater importance for lung
densitometry studies of conditions such as pulmonary fibrosis,
which produce changes in tissues of greater density than those
affected by the emphysematous process (19).
Good correlation has been reported between quantitative CT
and other measures of emphysema both for the voxel index
method (6, 7, 20–24) and the percentile method (22). Gevenois
and coworkers have shown that the optimal VI threshold (using
a high-resolution protocol) is ⫺950 HU in cross-sectional studies
comparing CT with both macroscopic (20) and microscopic mor-
phometry (6). In the current study, several VI threshold values
were used because it was considered that sensitivity to the
changes of emphysema progression would be threshold depen-
dent and, although the correction method was expected to im-
prove variability, it was recognized that the effect could also be
to change the average VI threshold to one less sensitive than
⫺910 HU. There are limited data on the measurement of em-
Parr, Stoel, Stolk, et al.: Calibration in CT Lung Densitometry
TABLE 4. ANNUAL PROGRESSION RATES OF SINGLE IMAGE
VOXEL INDEX AT A THRESHOLD OF ⫺910 HU, ADJUSTED
FOR CHANGES IN AIR CALIBRATION WITH
DATA PREVIOUSLY REPORTED AND INCLUDED
FOR COMPARISON*
Median Adjusted Rate Mean Annual Rate Reported
Image† (IQR)† Previously (SE)‡
UZI 1.67 (–0.23 to 3.28) 2.82 (0.55)
LZI 0.61 (–1.63 to 2.72) 1.03 (0.56)
UZE 0.58 (–0.50 to 2.98) 1.67 (0.38)
LZE 1.17 (–0.29 to 3.85) 1.66 (0.77)
Definition of abbreviations: IQR ⫽ interquartile range; LZE ⫽ lower-zone expir-
atory; LZI ⫽ lower-zone inspiratory; SE ⫽ standard error; UZE ⫽ upper-zone
expiratory; UZI ⫽ upper-zone inspiratory.
* See Reference 8.
†
n ⫽ 57.
‡
n ⫽ 43.
physema progression by serial quantitative CT, but in a study
of 22 patients with severe AATD over 2–4 years Dirksen and
coworkers (4) showed that the optimum VI threshold for whole
lung analysis was ⫺930 HU. The data presented here on two
different single lung slices confirm that ⫺930 HU was able to
detect progression with a high degree of statistical significance,
as were the other thresholds. The threshold that appeared statis-
tically the best for progression using this high-resolution protocol
(as determined by the Z statistic) was ⫺950 HU, although the
practicality of validating this in a longitudinal study using pathol-
ogy is preclusive. Nevertheless, the good correlation of this
threshold with pathology in cross-sectional studies (6, 20) sug-
gests that the changes in our study (at least on inspiratory scans
that would not be influenced by changes in air trapping, as in
expiratory scans) also reflect progression of emphysema.
Notwithstanding the errors resulting from variability in sam-
pling and inspiratory level, the single-slice protocol employed
here is sensitive to emphysema progression and also limits radia-
tion exposure. Volume scanning protocols may prove to be more
reproducible and sensitive to progression and the development
of low radiation dose protocols, in combination with faster scan-
ning of the chest, currently facilitates the use of volume scans
in ongoing clinical trials. These newer protocols allow correction
for differences in inspiration level between scans, which may
improve consistency further in longitudinal studies and also en-
able the retrieval of single slices from whole lung image sets to
identify regions in the lung with a better response to treatment
for emphysema.
In summary, air calibration is crucial to the consistency of lung
densitometry studies in emphysema, where tissue is replaced by
air. Air number measurements obtained from patient images
are more relevant to lung densitometry than standard quality
assurance values and may be used in a simple correction process
that adjusts for any air calibration errors. Although this process
may also be applicable as the first step toward the standardiza-
tion of lung densitometry between centers, further studies are
needed. Details of calibration method are essential to the inter-
pretation of lung densitometry, and quality assurance procedures
for densitometry studies should include the technique of air
calibration described here. These data must be reported in lon-
gitudinal studies of disease progression and when direct com-
parison is made between densitometry performed on different
scanners to ensure validity.
889
Conflict of Interest Statement : D.G.P. does not have a financial relationship with
a commercial entity that has an interest in the subject of this manuscript; B.C.S.
does not have a financial relationship with a commercial entity that has an interest
in the subject of this manuscript; J.S. does not have a financial relationship with
a commercial entity that has an interest in the subject of this manuscript; P.G.N.
does not have a financial relationship with a commercial entity that has an interest
in the subject of this manuscript; R.A.S. has lectured widely for non-promotional
purposes for GlaxoSmithKline, Bayer, and Eli Lilly and acts on Advisory Boards for
AstraZeneca (AZ) and Baxter and has two non-commercial research grants—one
from AZ (£125,000 since 2002) and one from Bayer (£500,000 per annum since
1996).
Acknowledgment : The authors thank KCARE (Kings College Hospital, London)
for the loan of the lung phantom and Derek Tarrant for technical support with
the use of the Marconi MxView scanner. R. A. Stockley and J. Stolk are members
of AIR (the Alpha-1 International Registry, www.aatregistry.org) and thank the
other council members of AIR for valuable discussions on the results of this
study.
References
1. Burrows B. A clinical trial of efficacy of antiproteolytic therapy: can it
be done? Am Rev Respir Dis 1983;127:S42–S43.
2. Croxton TL, Weinmann GG, Senior RM, Wise RA, Crapo JD, Buist
AS. Clinical research in chronic obstructive pulmonary disease: needs
and opportunities. Am J Respir Crit Care Med 2003;167:1142–1149.
3. Snider GL, Kleinerman J, Thurlbeck WM, Bengali ZH. The definition
of emphysema: report of a National Heart, Lung, and Blood Institute,
Division of Lung Diseases Workshop. Am Rev Respir Dis 1985;132:
182–185.
4. Dirksen A, Friis M, Olesen KP, Skovgaard LT, Sorensen K. Progress of
emphysema in severe ␣1-antitrypsin deficiency as assessed by annual
CT. Acta Radiol 1997;38:826–832.
5. Gevenois PA, Zanen J, de Maertelaer V, De Vuyst P, Dumortier P,
Yernault JC. Macroscopic assessment of pulmonary emphysema by
image analysis. J Clin Pathol 1995;48:318–322.
6. Gevenois PA, De Vuyst P, de Maertelaer V, Zanen J, Jacobovitz D, Cosio
MG, Yernault JC. Comparison of computed density and microscopic
morphometry in pulmonary emphysema. Am J Respir Crit Care Med
1996;154:187–192.
7. Müller NL, Staples CA, Miller RR, Abboud RT. Density mask: an objec-
tive method to quantitate emphysema using computed tomography.
Chest 1988;94:782–787.
8. Dowson LJ, Guest PJ, Stockley RA. Longitudinal changes in physiologi-
cal, radiological, and health status measurements in ␣1-antitrypsin de-
ficiency and factors associated with decline. Am J Respir Crit Care
Med 2001;164:1805–1809.
9. Gierada DS, Yusen RD, Pilgram TK, Crouch L, Slone RM, Bae KT,
Lefrak SS, Cooper JD. Repeatability of quantitative CT indexes of
emphysema in patients evaluated for lung volume reduction surgery.
Radiology 2001;220:448–454.
10. Kemerink GJ, Lamers RJ, Thelissen GR, van Engelshoven JM. CT densi-
tometry of the lungs: scanner performance. J Comput Assist Tomogr
1996;20:24–33.
11. Stoel BC, Vrooman HA, Stolk J, Reiber JH. Sources of error in lung
densitometry with CT. Invest Radiol 1999;34:303–309.
12. Kemerink GJ, Lamers RJ, Thelissen GR, van Engelshoven JM. Scanner
conformity in CT densitometry of the lungs. Radiology 1995;197:749–
752.
13. Morgan MD. Detection and quantification of pulmonary emphysema by
computed tomography: a window of opportunity. Thorax 1992;47:1001–
1004.
14. Dirksen A, Dijkman JH, Madsen F, Stoel BC, Hutchison DC, Ulrik CS,
Skovgaard LT, Kok-Jensen A, Rudolphus A, Seersholm N, et al. A
randomized clinical trial of ␣1-antitrypsin augmentation therapy. Am
J Respir Crit Care Med 1999;160:1468–1472.
15. Parr DG, Stoel BC, Stolk J, Stockley RA. Lung densitometry: achieving
standardization by overcoming variability in scanner performance [ab-
stract]. Am J Respir Crit Care Med 2004;169:A881.
16. Dowson LJ, Newall C, Guest PJ, Hill SL, Stockley RA. Exercise capacity
predicts health status in ␣1-antitrypsin deficiency. Am J Respir Crit
Care Med 2001;163:936–941.
17. Stolk J, Ng WH, Bakker ME, Reiber JH, Rabe KF, Putter H, Stoel BC.
Correlation between annual change in health status and computer
tomography derived lung density in subjects with ␣1-antitrypsin defi-
ciency. Thorax 2003;58:1027–1030.
18. Stoel BC, Bakker ME, Stolk J, Dirksen A, Stockley RA, Piitulainen E,
Russi EW, Reiber JH. Comparison of the sensitivities of 5 different
890 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 170 2004
computed tomography scanners for the assessment of the progression
of pulmonary emphysema: a phantom study. Invest Radiol 2004;39:1–7.
19. Best AC, Lynch AM, Bozic CM, Miller D, Grunwald GK, Lynch DA.
Quantitative CT indexes in idiopathic pulmonary fibrosis: relationship
with physiologic impairment. Radiology 2003;228:407–414.
20. Gevenois PA, de Maertelaer V, De Vuyst P, Zanen J, Yernault JC.
Comparison of computed density and macroscopic morphometry in
pulmonary emphysema. Am J Respir Crit Care Med 1995;152:653–657.
21. Gevenois PA, De Vuyst P, Sy M, Scillia P, Chaminade L, de Maertelaer
V, Zanen J, Yernault JC. Pulmonary emphysema: quantitative CT
during expiration. Radiology 1996;199:825–829.
22. Gould GA, MacNee W, McLean A, Warren PM, Redpath A, Best JJ,
Lamb D, Flenley DC. CT measurements of lung density in life can
quantitate distal airspace enlargement: an essential defining feature
of human emphysema. Am Rev Respir Dis 1988;137:380–392.
23. Hayhurst MD, MacNee W, Flenley DC, Wright D, McLean A, Lamb
D, Wightman AJ, Best J. Diagnosis of pulmonary emphysema by com-
puterised tomography. Lancet 1984;2:320–322.
24. Kuwano K, Matsuba K, Ikeda T, Murakami J, Araki A, Nishitani H,
Ishida T, Yasumoto K, Shigematsu N. The diagnosis of mild emphy-
sema: correlation of computed tomography and pathology scores. Am
Rev Respir Dis 1990;141:169–178.