| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 Department of Respiratory Medicine, Gentofte University Hospital, Copenhagen, Denmark
Correspondence and requests for reprints should be addressed to Asger Dirksen, M.D., Department of Respiratory Medicine, Gentofte University Hospital, Copenhagen, Denmark. E-mail: adi{at}dadlnet.dk
ABSTRACT
Emphysema is a major constituent of lung pathology in chronic obstructive pulmonary disease, and emphysema is characterized by loss of lung tissue. Computed tomography (CT) generates detailed information on tissue densities, and lung density by CT has the potential of being both more sensitive and specific for monitoring emphysema in clinical trials than more traditional outcome measures such as pulmonary function tests. However, CT measurements of lung density are subject to many sources of noise, such as variations in the scanning procedure (technical noise) and variations in patient performance (biological noise). A multi-detector array scanner and a low-dose volumetric scanning protocol are recommendable, and images should be reconstructed using a soft filter. In longitudinal studies the percentile density (PD) is the most reproducible densitometric parameter, and confounding due to variations in inspiratory level can be adjusted for by a physiologic model that assumes that the lung behaves like a sponge. Work remains to be done to standardize and validate CT lung density, before it can become established as the primary outcome measure in clinical trials of new treatments for chronic obstructive pulmonary disease.
Key Words: alpha-1 antitrypsin deficiency • chronic obstructive pulmonary disease • computed tomography • pulmonary emphysema
EMPHYSEMA AS OUTCOME IN CHRONIC OBSTRUCTIVE PULMONARY DISEASE
Emphysema is a major constituent of lung pathology in chronic obstructive pulmonary disease (COPD) and is the major determinant of both airflow obstruction and impaired diffusion capacity in COPD (1). Emphysema is characterized by progressive loss of lung tissue, which is replaced by air, resulting in a decrease in the physical density of the lung. This decrease can be assessed by repeated measurements of lung density by computed tomography (CT) (i.e., lung densitometry) (2). CT is the imaging method of choice to assess the extent of emphysema in life, and studies have shown that objective quantitation by CT has good correlation with the pathologic extent of emphysema (3, 4). Findings from a randomized controlled trial of the protective effect of 
1-antitrypsin replacement therapy in patients with deficiency suggested that lung densitometry derived from annual CT scans is more sensitive than FEV1 and the carbon monoxide transfer coefficient (KCO) in monitoring the progression of the disease (5).
LONGITUDINAL VERSUS CROSS-SECTIONAL STUDY DESIGN
For any outcome in clinical trials, it is appropriate from a statistical point of view to distinguish between variation between subjects (inter-individual variation) and within subjects (intra-individual variation), that is, variation in lung density between repeat scans of the same individual. The statistical advantage of longitudinal studies lays in the possibility of eliminating the inter-individual variability at baseline when comparing the outcomes in various treatment groups (6). However, a longitudinal study is more laborious than the corresponding cross-sectional study, and therefore, longitudinal studies are recommendable only in situations where the intra-individual variation is small as compared with the inter-individual variation, and this is usually the case when using CT lung density as an outcome in clinical trials (5). Furthermore, in longitudinal studies it is always essential for the statistical power of the trial to keep the intra-individual variation as low as possible, that is, to minimize the noise of repeated measurements (6).
Various Densitometric Parameters
Based on the frequency distribution histogram of voxel attenuation values of the whole lung, many different variables have been used in lung densitometry of which the emphysema percentage and the percentile density are the most reproducible (7, 8). The emphysema percentage (EP) is based on the assumption that voxels with densities below a certain threshold (often –910 Hounsfield Units [HU]) represent emphysema, and the emphysema percentage is defined as the percentage of the total lung volume that contains voxels with densities below this threshold. The percentile density (PD) is defined as the value (HU) at which a certain percentage (usually 15%) of the voxels in the frequency distribution histogram have a lower density, and may be expressed as g/L by the simple addition of 1,000 to the Hounsfield value of the percentile density.
Linearity of Time-Trend of Various Densitometric Parameters
The effect of various treatments are usually compared by "slope" analysis, that is, change over time of an outcome variable is calculated as a slope by linear regression analysis, and mean slopes are compared between treatment groups (5). For some variables (such as EP), slope is strongly correlated to lung density at baseline (Figure 1). This is a problem when using EP as an outcome measure, because reliable method for eliminating this important source of statistical noise is not available (9). Baseline lung density has less influence on the slope of PD (8, 10). Furthermore, for PD the slope is the same for a wide range of percentages (from 10% to 30%), whereas for EP results vary a lot depending on the choice of threshold (from –950 HU to –850 HU) (9). For these reasons the PD 15% was recommended as the best densitometric parameter in an expert review (11).
|
Calibration is mandatory to ensure that Hounsfield numbers for air and water are measured at –1,000 HU and 0 HU, respectively (12). Lung density values lie between these two points and, if linearity is assumed, adjustment of air and water numbers should ensure validity of lung values. It is practical to distinguish between scanner (external) calibration and image (internal) calibration.
Scanner air calibration (with an empty gantry) should be performed according to the scanner manufacturers' instructions within 3 hours of the first patient scan, and every 3 hours during scanning lists. Scanner water calibration is usually performed by the manufacturers (using the manufacturers' water phantom) at least every 3 months using the clinical scan protocol.
Measurement of blood density in the subdiaphragmatic descending thoracic aorta and air density in the preventral area may be used for internal calibration of image sequences. Thus, internal calibration allows adjustment of densitometric values by rescaling lung segmentation according to measurements obtained from the image series. Reliable external calibration may reduce the need for internal calibration. However, the opposite is probably not true; internal calibration may only to some extent correct for insufficient external calibration.
THE SCANNING PROTOCOL
CT measurements of lung density are subject to many sources of noise due to variations in the scanning procedure (technical noise) and variations in patient performance (biological noise) (Table 1). Scan acquisition parameters should be standardized, and radiation per CT scan should be kept low, that is, around or below 1 mSv. A low-dose volumetric scanning protocol should be used, and with the aim of minimizing breathing artifacts multi-detector array CT scans should be performed in the supine position and within 10 seconds of breath hold. Patients should be told to inspire as close to total lung capacity as possible to improve reproducibility and reduce low-density artifacts arising from air-trapping. Scanning should be performed in a caudo-cranial direction, to reduce artifacts arising from diaphragm movement, and with the arms raised above the head to reduce densitometric artifact arising from X-ray spectral changes and beam scatter due to juxtaposition of dense tissue adjacent to the lungs. No contrast medium should be used. Some investigators recommend the use of inhaled bronchodilator therapy 30 minutes before scanning and the performance of three deep inspiratory maneuvers immediately before scanning to ensure standardized expansion and ventilation of basal areas that are susceptible to atelectatic changes, but no data are available to support these recommendations.
|
Image reconstruction algorithms should be performed according to a standardized protocol for each CT scanner, using a soft filter (defined by the type of the CT scanner). Quantum mechanics implies a trade-off between spatial and contrast resolution (13), and it has been shown that the combination of low radiation dose, hard reconstruction algorithm, and thin slices (1 mm) results in overestimation of the extent of emphysema (14). Using current scanning techniques, a slice thickness between 3 and 5 mm, and the use of a soft reconstruction filter is probably the optimal combination (14).
SOFTWARE FOR IMAGE ANALYSIS
Although an increasing number of semi-automated software for image analysis is commercially available, many scientists still use homemade in-house software with its inherent limitations. These programs delineate lung parenchyma by automatic lung segmentation using a method known as the "seeded-region growing technique" and defining a threshold between lung and extrapulmonary tissue of typically –400 HU (15). For instance, the trachea is located manually, and the program automatically identifies low-density voxels below the chosen threshold value that are contiguous with the tracheal origin. Exclusion of the trachea between the point of seeding and the carina can occur automatically after lung segmentation, and subsequently using more sophisticated methods the bronchial tree can be segmented almost down to the bronchiolar level (16). Some programs use smoothing techniques in the hilar region for segmenting the lung, which reduces the complexity of the line that defines the lung, but it may impair the reproducibility of lung segmentation in repeat scans (17–19). Many programs allow for manual correction of errors in contour detection, such as inclusion of air in the esophagus or bowel.
BIOLOGICAL NOISE IS MAINLY DUE TO VARIATION IN INSPIRATORY LEVEL
By paying careful attention to the scanning procedures, it should be possible to reduce technical noise to a minimum, and therefore biological noise—that is, variations in subject performance—is usually the most important source of noise. Thus, the level of inspiration during scan acquisition is recognized to influence lung density and reduces the reproducibility of CT lung densitometry, and in longitudinal studies, it is necessary to remove this confounder to detect changes in lung density due to disease progression (20). In co-operative patients, breath-holding at full inspiration is the most reproducible lung volume, resulting in the lowest variation in lung density between consecutive scans (21). However, it is well known that spirometric test results fluctuate even during the same day, and spirometric gating during scanning does not improve the accuracy of the results (22). Careful breathing instructions during scanning, in a manner similar to the lung function laboratory, are strongly recommended. However, even then it is essential subsequently to adjust lung density measurements for variations in the total lung volume.
VOLUME ADJUSTMENT OF LUNG DENSITY MEASUREMENTS
Two different methods have been proposed to standardize percentile densities for variation in total lung volume (TLV) measured from CT. One method is statistical (7), whereas the other method is based on a physiologic model (9). The statistical method simply includes TLV as a covariate in the statistical model with lung density as the (dependent) endpoint and time, treatment effect, and other confounders as (independent) predictors.
SPONGE-MODEL FOR VOLUME ADJUSTMENT OF PERCENTILE DENSITIES
The physiologic model is based on several assumptions. From the definition of density as weight per volume, it follows that the product of density and volume is equal to weight (weight = volume x density). Furthermore, it is assumed that variations in inspiratory level have no influence on the total weight of the lung that stays constant, which means that, for instance, halving the volume should double the density. In other words, the physiologic model assumes that the lung behaves like a sponge, and therefore, it is also referred to as the "sponge-model." The assumption of constant lung weight may seem incompatible with the fact that the generation of negative intrathoracic pressure required for inspiration will necessarily be accompanied by a variable increase in blood flow to the lung. However, most of the blood flow into the lung is intercepted by large capacitance veins, and with a threshold for the soft tissue–lung interface of –400 HU, large vessels are excluded from "lung" by the region-growing algorithm. It has, therefore, been argued that this physiologic effect should not influence the weight of the lung as calculated from the CT scans, and this hypothesis is supported by recent clinical data (9) that showed no correlation between TLV and the total weight of the lung (Figure 2). Finally, it is assumed that the lung expands uniformly and ventilation is uniformly distributed throughout the lung, and this is supported by old studies in animals at total lung capacity (TLC) published in the Journal of Applied Physiology (23, 24).
|
 |
 |
Physiologic adjustment has the advantage of being intuitively meaningful (the lung behaves like a sponge), and each density value can be adjusted for volume independent of density measurements from other scans. The statistical adjustment is easy to perform, but adjustment will depend on all the density measurements that are included in the statistical analysis. Furthermore, the physiologic interpretation of the adjustment is not so obvious, and in this respect it is worrisome that the statistical model will predict negative lung densities beyond a certain lung volume.
OTHER SOURCES OF BIOLOGICAL NOISE
Apart from variations in inspiratory level, biological noise due to co-morbidities such as pneumonia and congestive heart failure may also heavily influence lung density, and therefore it is strongly recommended that visual inspection of images be performed before study analysis. Preferably this should be done by a blinded review panel of experts with full access to clinical information about the patients.
CONCLUSIONS
In conclusion, CT lung density has the potential of being both more sensitive and specific for emphysema than more traditional outcome measures such as pulmonary function tests and health-related quality of life, but still some work needs to be done to standardize and validate CT lung density before it can become established as the primary outcome measure in clinical trials of new treatments for COPD (26).
FOOTNOTES
Conflict of Interest Statement: A.D. has participated as a speaker in scientific meetings organized and financed by Bayer and Talecris. He received $200,000 in 2005 and 2006 from Bayer as research grants for participating as principal investigator in a multi-center randomized clinical trial.
(Received in original form April 3, 2008; accepted in final form July 13, 2008)
REFERENCES
This article has been cited by other articles:
|
|
 |
|
  H. O. Coxson, J. Mayo, S. Lam, G. Santyr, G. Parraga, and D. D. Sin New and Current Clinical Imaging Techniques to Study Chronic Obstructive Pulmonary Disease Am. J. Respir. Crit. Care Med., October 1, 2009; 180(7): 588 - 597. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
