| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Thoracic Imaging |
1 From the Departments of Radiology (A.C.B., A.M.L., D.M., D.A.L.) and Preventive Medicine and Biometrics (G.K.G.), University of Colorado Health Sciences Center, 3025 E 11th Ave, Denver, CO 80262; and Biogen, Cambridge, Mass (C.M.B.). From the 2000 RSNA scientific assembly. Received March 21, 2002; revision requested June 4; final revision received December 19; accepted February 24, 2003. Supported by Biogen. Address correspondence to A.C.B. (e-mail: alan.best@uchsc.edu).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS DISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: The authors analyzed baseline digital thin-section CT data from 144 patients with IPF who enrolled in a double-blind placebo-controlled clinical effectiveness trial of interferon beta 1a in the treatment of IPF. All patients underwent thin-section CT in the supine position at full inspiration. The lungs were isolated by using a semiautomated thresholding technique, with an upper threshold of -200 HU. An attenuation correction algorithm was used. Pulmonary function tests (PFTs) included forced vital capacity, total lung capacity, forced expiratory volume in 1 second, and diffusing lung capacity. Univariate and multiple correlation and regression statistical analyses were used to determine relationships between histogram features and results of PFTs.
RESULTS: Moderate correlations existed between histogram features and PFT results. Kurtosis showed the greatest degree of correlation with physiologic abnormality (r = 0.53, P < .01). Strength of correlation increased with exclusion of suboptimal scans but did not change significantly after application of an attenuation correction algorithm. Attenuations for lungs, gas, and soft tissue varied considerably between scanner manufacturers. Kurtosis alone provided predictions of pulmonary function that were virtually as good as those from all histogram features combined.
CONCLUSION: Thin-section CT histograms of the lungs were found to correlate with results of PFTs in patients with IPF, which supports the claim that histogram features can be used as valid indexes of IPF in a multiinstitutional nonspirometrically controlled study.
© RSNA, 2003
Index terms: Computed tomography (CT), comparative studies, 60.12118 • Computed tomography (CT), image quality, 60.12118 • Computed tomography (CT), technology, 60.12118 • Computed tomography (CT), thin-section, 60.12118 • Lung, CT, 60.12118 • Lung, fibrosis, 60.792 • Lung, ventilation
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS DISCUSSIONREFERENCES |
|---|
Computer-derived indexes, such as mean lung attenuation, skewness, and kurtosis, can be obtained from frequency histograms of thin-section CT scans of the lung (1). Mean lung attenuation represents the average global attenuation value of the lung. Skewness describes the degree of asymmetry of a histogram; a histogram with a long tail to the right has a positive skewness value, and a perfectly symmetric distribution has a skewness value of zero. Kurtosis describes how sharply peaked a histogram is; a histogram that is more peaked than a normal distribution has a positive kurtosis value, and a normal distribution has a kurtosis of zero. These three histogram features are quantitative and reproducible with minimal user intervention and can be obtained from digital thin-section CT scans by using an automated computer thresholding technique. Thus, these indexes are attractive for scoring disease extent in patients with IPF. In healthy subjects, the first-order histogram of CT attenuation is sharply peaked (kurtotic) and skewed to the left in comparison to the Gaussian normal distribution (Fig 1). Histograms from patients with IPF are less skewed, less kurtotic, and have increased mean lung attenuation compared with those from patients with normal lungs (2–4) (Fig 2a, 2b). This is thought to result from an increase in soft tissue and a decrease in gas.
|
|
|
|
|
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS DISCUSSIONREFERENCES |
|---|
The data used in the present study came from baseline CT scans obtained in 144 subjects. Examinations were performed at 30 participating hospitals in the United States and Canada and included the administration of PFTs and acquisition of thin-section CT scans of the lungs. Major inclusion criteria consisted of a confirmed diagnosis of IPF and disease progression while being treated with a steroid or cytotoxic agent. Major exclusion criteria were asbestos exposure or prior environmental exposure to any drug or agent known to cause pulmonary fibrosis, such as nitrofurantoin. Patients were excluded from analysis if their CT scans were rated as suboptimal for thin-section technique, acquisition, level of inspiration, or lack of respiratory motion. Institutional review board approval and patient informed consent were obtained at each participating institution for the double-blind placebo-controlled clinical study. The present study was exempt from review by the Colorado Multiple Institutional Review Board, or COMIRB. Informed consent was not required because the research involved the study of existing data, which were assessed and recorded by the authors in such a manner that subjects could not be identified, either directly or by means of identifiers linked to the subjects.
PFTs were used to measure forced vital capacity (FVC), forced expiratory volume in 1 second (FEV1), total lung capacity (TLC), and diffusing capacity. The mean time between PFT administration and thin-section CT scanning was 7 days ± 11 (range, 0–50 days). These tests were performed at each center by using commonly accepted measurement techniques. Results were expressed as a percentage of predicted performance by using standards developed by Crapo et al (6) and Crapo and Morris (7) for FVC, FEV1, and diffusing lung capacity and by using the Goldman and Becklake (8) formula for TLC.
Determination of Mean Lung Attenuation, Skewness, and Kurtosis
The study protocol called for acquisition of a series of scans of the entire lung, with 1-mm section thickness and 2-cm spacing. Subjects were placed in the supine position and were imaged at full inspiration. Supine positioning allowed for standardization of lung expansion, which is dependent on patient position (7,9). Subjects were asked to take a deep breath and hold it prior to acquisition of each scan. Scans were reconstructed by using a thin-section algorithm. Participating hospitals mailed digital thin-section CT scans along with hard-copy images to our institution. Digital scans were mailed on digital audio tapes, compact disks, or optical disks in digital imaging and communications in medicine, or DICOM, format. Hard-copy images were assessed visually by a radiologist (D.A.L.) for scan quality, including the adequacy of inspiration.
Once digital images were received at the central reading facility, they were transferred from the storage media to a DataStor P5-166 (Garden Grove, Calif) personal computer with a 166-MHz Pentium processor, with which all subsequent segmentation and image analysis was performed. Images were segmented by using a program written in Interactive Data Language (IDL; Research Systems, Boulder, Colo). This program uses a semiautomated thresholding technique to isolate the lungs from other tissues and structures and selects all pixels between -200 and -1,000 HU. These limits are similar to those used in a prior study (10). Since the lower limit of the Hounsfield unit scale is -1,024 HU with some scanners, the program linearly corrected the scale to a standard Hounsfield unit scale of 0 to -1,000 HU. Minimal user intervention by one author (A.C.B.) was required to exclude those nonlung structures that satisfied the threshold criteria, such as the trachea, blood vessels, and large bronchi near the hilum (Fig 2c). Values for skewness, kurtosis, and mean lung attenuation were calculated at every image level for the entire lung in every patient.
Comparability of Scanners
Digital data were acquired at 30 institutions. Five manufacturers were included in this study: GE Medical Systems, Waukesha, Wis; Imatron, San Francisco, Calif; Philips Medical Systems, Andover, Mass; Picker, Cleveland, Ohio; and Siemens, Malvern, Pa. Federal law requires that CT scanners be calibrated periodically and checked for quality; however, variations between scanners still occur. To ensure comparability of mean lung attenuation measurements obtained with the use of different scanners, a plastic CT lung phantom (Computerized Imaging Reference Systems, Norfolk, Va) that contained lung-equivalent and tissue-equivalent material was mailed to each site to be scanned. After the phantom was scanned, it was returned to the central reading facility with the digital images. The values for gas, lung, and soft-tissue attenuation at each site were calculated automatically and recorded. Because of corruption of data and hardware incompatibilities, phantom scans were obtained at only 12 of the 30 sites.
As an alternate method to ensure comparability, the attenuations of myocardial tissue and tracheal gas were measured directly from scans obtained in every subject. Five regions of interest (50–90 mm2) were selected manually by one author (A.C.B.) in both the trachea and the myocardium. Mean attenuation values were calculated for tracheal gas and myocardium in each patient. The measured mean lung attenuation for each patient was multiplied by the following correction factor to obtain the corrected mean lung attenuation: Correction factor = (heart attenuation - tracheal gas attenuation)/1,040.
In this equation, the value 1,040 represents the difference between the expected value of myocardial attenuation (40 HU) and that of gas (-1,000 HU). The value of 40 HU was chosen arbitrarily as a typical value for soft-tissue attenuation. It was believed that this method of mean lung attenuation correction would compensate for variability between scanners. The correction does not alter skewness or kurtosis because these quantities are not affected by a multiplicative correction (they are scale invariant).
Statistical Analysis
Statistical analyses were performed by using JMP (SAS Institute, Cary, NC) and S-plus (Insightful, Seattle, Wash) software. Pearson correlation coefficients (r values) were used to quantify the relationships between each of the four histogram features (skewness, kurtosis, mean lung attenuation, and corrected mean lung attenuation) and each of the four PFTs (diffusing lung capacity, TLC, FEV1, and FVC). Sensitivity of these analyses to a single manufacturer (manufacturer 2) and to suboptimal scans was assessed by means of repeat analysis after omitting all scans from manufacturer 2 and the suboptimal scans.
There is no well-defined threshold for acceptability of strength of correlation. A high degree of correlation may indicate that histogram features can provide a useful indicator of disease progression. Multiple regression analysis was used to determine predictive values of histogram features alone and together. For assessment of each of the four physiologic outcomes separately (FVC, FEV1, TLC, and diffusing lung capacity), four regression models were estimated: Three with each histogram feature as a single predictor (mean lung attenuation, skewness, and kurtosis), and one with all three predictors together. The multiple regression with all three predictors together took into account the strong correlations among the histogram features.
To assess sensitivity of results to manufacturers, these regressions were performed by using scans from manufacturer 1, manufacturer 2, and all other manufacturers combined. The predictive performance of each regression was summarized by means of the R2 value and the residual SD (
). Higher R2 and lower values indicate better model predictions. It is believed that residual prediction error is more useful than R2 for the indication of predictive ability, since R2 depends on the spread of the values of PFTs, while does not. Finally, backward selection with an exclusion P value of .05 was used to obtain models for each physiologic outcome separately by using fewer histographic predictors but having essentially equal predictive ability to that of the model with all variables. The resulting prediction equations are given for possible use by other clinicians or investigators. The backward selection analyses involved use of data from all manufacturers combined, since in practical application of these prediction equations, the manufacturer may not be one of those used in the present study.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS DISCUSSIONREFERENCES |
|---|
|
|
|
|
|
As expected, correlation coefficients for results between PFTs were high. Within this group, the correlation between FVC and FEV1 was greatest (r = 0.94). Similarly, correlation coefficients between histogram features were also high (r = 0.96 between mean lung attenuation and corrected mean lung attenuation; r = 0.95 between skewness and kurtosis).
Exclusion of Suboptimal Scans
Of the 144 patient scans used in this study, 34 scans (24%) were rated as suboptimal by one radiologist (D.A.L.), usually because of the presence of mild or moderate motion artifact, which was often associated with hypoventilation. Exclusion of these suboptimal scans moderately increased 14 of the 16 correlation coefficients between physiologic measures and histogram features, as shown in Table 4. The greatest degree of increase was that between skewness and FVC (from 0.47 to 0.55). One of the correlation coefficients (diffusing lung capacity and kurtosis) did not change, while TLC and corrected mean lung attenuation decreased from -0.35 to -0.33. The mean value of mean lung attenuation for all excluded scans was -689.3 HU ± 62.2, which was similar to that for all included scans. There were no significant differences in physiologic values between patients with excluded scans and those with included scans.
Analysis according to Scanner Manufacturer
The exclusion of all scans obtained with CT scanners made by manufacturer 2 resulted in a moderate increase in correlation coefficients in every category, as shown in Table 5. This may be due in part to an increased number of suboptimal scans obtained with these machines. There was a significantly higher percentage of suboptimal scans obtained with these scanners (P = .007). The number of scans with optimal quality and those with suboptimal quality from each manufacturer are given in Table 1.
The attenuation values for phantom scans are shown in Table 6. For all phantom scans, mean lung attenuation was -793.9 HU, mean gas attenuation was -985.0 HU, and mean myocardial attenuation was 57.1 HU. Myocardial attenuation measurements from phantoms were somewhat higher than those from patients scanned at the same institutions. Thus, correction factors derived from phantom scans were slightly higher than those derived from patients. For the mean values of lung, gas, and myocardial attenuations, manufacturers 1, 2, and 3 were within 20 HU of each other. The SDs of measured phantom attenuations by manufacturer are also given in Table 6.
|
|
|
|
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS DISCUSSIONREFERENCES |
|---|
As shown in Table 2, the patient population in this study had higher mean lung attenuation, less skewness, and less kurtosis than those in a study of patients with IPF by Hartley et al (1), which suggests that the population in the present study had more advanced IPF than did those in the prior study. As expected, the mean lung attenuation values from these IPF populations are higher than those from healthy populations. The results of two prior studies of healthy subjects exhibited mean lung attenuation values of -798 HU (11) and -819 HU (12). These two study populations with six and 24 participants, respectively, consisted of healthy volunteers imaged with 50% spirometric gating. In the present study, patients were instructed to take a deep breath and hold it. This may have resulted in deeper inspiration than that obtained in the two prior study populations imaged with 50% spirometric gating. Deeper inspiration would be expected to decrease mean lung attenuation, increase skewness, and increase kurtosis.
The correlation coefficients between histogram features and physiologic impairment in the present study were moderately lower than those in prior studies (1,3,11) (Table 2). This is thought to be a result of the multiinstitutional nature of the study and the lack of spirometric gating.
A mean lung attenuation correction technique was used in an attempt to ensure comparability of the different thin-section CT scanners used in this study. It is interesting that mean lung attenuation correction had little effect on values of correlation coefficients. It is also interesting that the CT scanners consistently yielded a higher value of gas attenuation than the known value of -1,000 HU. This may be due to the use of reconstruction algorithms and artifact from tissues surrounding the trachea.
The 144 scans included in the study were rated for lack of respiratory motion artifact. Thirty-four of these scans had motion artifact. As expected, exclusion of these scans moderately increased the values of correlation coefficients. Therefore, scans with even mild motion artifact should be excluded from digital analysis. The mean values of mean lung attenuation and results of physiologic tests from excluded scans were similar to those from the included scans, which suggests similarity in disease severity. Thus, it is believed that exclusion of suboptimal scans did not introduce a bias.
Exclusion of scans obtained with manufacturer 2 scanners improved correlation coefficients, and it is thought that this resulted in part from the significantly higher incidence (18 of 47 scans, 38%) of suboptimal scans with these scanners. The findings on phantom scans and the wide variation in attenuation of tracheal gas and myocardial tissue indicate that substantial variations exist between scanner manufacturers. These variations may be due to inconsistent calibration or to variations in beam-hardening corrections or reconstruction algorithms. Phantom measurements of soft-tissue–equivalent attenuation were slightly higher than myocardial attenuation measurements in patients scanned at the same institutions. This is most likely due to higher density of soft-tissue–equivalent plastic in the phantom compared with actual myocardial tissue. For future studies of this type, we recommend a greater emphasis on calibration and perhaps the use of a smoothing rather than an edge-enhancing reconstruction algorithm.
There are many potential problems and complications associated with a multicenter trial such as this one, which may have resulted in a decrease of correlation values relative to those obtained in single-center trials. Standardization of protocols is essential to ensure comparability of data from each of the 30 participating sites. The multiple steps involved in the longitudinal process of image acquisition, transfer, and analysis provided many opportunities for problems, including incompatible equipment, lack of adherence to CT protocols, and variations in technologist instructions to the patient. Despite efforts at standardization, interscanner variability undoubtedly played a substantial role in this study, including differing CT calibrations and reconstruction algorithms. Other deviations from study protocol may have occurred in the administration of PFTs.
Lung attenuation is a function of level of inspiration. Steps can be taken to standardize the level of patient inspiration, such as the use of a handheld spirometer or a pneumotachometer. The equipment for spirometric gating is not readily commercially available, and its use in this study was prevented by the logistics of providing such equipment to each site. However, variability due to lack of gating was thought to be small compared with the magnitude of change in mean lung attenuation associated with physiologic dysfunction. Additionally, there was no correction for the effect of lung volume on mean lung attenuation in this study. For future studies, lung volume can be computed directly from spiral CT images (13), which may eliminate the need for spirometric gating.
PFTs are not a perfect measure of the underlying disease state. Thus, both the PFT results and histogram features may have a high correlation with the progression of disease (for which a perfect quantitative descriptor may not exist) yet have only moderate correlation with each other. The advantage of using CT, rather than physiologic testing, for quantification of lung disease is that CT offers direct visualization of the morphologic extent of disease and potentially offers a direct measurement of the amount of soft tissue present in the lung.
Despite significant interscanner variation, measurements of mean lung attenuation, skewness, and kurtosis from thin-section CT histograms of the lungs were found to correlate with results of PFTs in patients with IPF, supporting the claim that CT histogram features can be used as valid indexes of IPF in a multiinstitutional nonspirometrically controlled study. Our results were consistent with those in a prior study (4). The magnitude of correlation was greatest between kurtosis and FVC, and kurtosis was the histographic predictor that correlated most highly with all four PFTs. Predictions derived by using all three histographic variables were slightly, but not substantially, better than those derived by using kurtosis alone. We found a moderate increase in correlation of histogram features with PFT results after the exclusion of CT images with motion artifact and found no significant change after an attenuation correction algorithm was applied.
The measurements of mean lung attenuation, skewness, and kurtosis used in the present study are relatively insensitive to textural changes, such as ground-glass abnormality, reticular abnormality, and honeycombing. More sophisticated image analysis techniques, such as fractal analysis (14) and the adaptive multiple feature method (15), may be more helpful in the quantification of disease extent. We plan to apply texture-based analysis to this data set in the near future.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Author contributions: Guarantor of integrity of entire study, A.C.B.; study concepts and design, all authors; literature research, A.C.B., A.M.L., C.M.B., D.A.L.; clinical studies, C.M.B., D.A.L.; experimental studies, A.C.B., D.A.L.; data acquisition, A.C.B., D.M., D.A.L.; data analysis/interpretation, A.C.B., D.M., G.K.G., D.A.L.; statistical analysis, A.C.B., D.M., G.K.G., D.A.L.; manuscript preparation, A.C.B., G.K.G., D.A.L.; manuscript definition of intellectual content and editing, A.C.B., A.M.L., G.K.G., D.A.L.; manuscript revision/review and final version approval, all authors.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS DISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  A. Madani, A. Van Muylem, V. de Maertelaer, J. Zanen, and P. A. Gevenois Pulmonary Emphysema: Size Distribution of Emphysematous Spaces on Multidetector CT Images--Comparison with Macroscopic and Microscopic Morphometry Radiology, September 1, 2008; 248(3): 1036 - 1041. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. C. Best, J. Meng, A. M. Lynch, C. M. Bozic, D. Miller, G. K. Grunwald, and D. A. Lynch Idiopathic Pulmonary Fibrosis: Physiologic Tests, Quantitative CT Indexes, and CT Visual Scores as Predictors of Mortality Radiology, March 1, 2008; 246(3): 935 - 940. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. A. Lynch Quantitative CT of Fibrotic Interstitial Lung Disease Chest, March 1, 2007; 131(3): 643 - 644. [Full Text] [PDF]
|
|
|
|
|
|
G. Camiciottoli, I. Orlandi, M. Bartolucci, E. Meoni, F. Nacci, S. Diciotti, C. Barcaroli, M. L. Conforti, M. Pistolesi, M. Matucci-Cerinic, et al. Lung CT Densitometry in Systemic Sclerosis: Correlation With Lung Function, Exercise Testing, and Quality of Life Chest, March 1, 2007; 131(3): 672 - 681. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Matsuoka, Y. Kurihara, K. Yagihashi, H. Niimi, and Y. Nakajima Quantification of Thin-Section CT Lung Attenuation in Acute Pulmonary Embolism: Correlations with Arterial Blood Gas Levels and CT Angiography. Am. J. Roentgenol., May 1, 2006; 186(5): 1272 - 1279. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Misumi and D. A. Lynch Idiopathic pulmonary fibrosis/usual interstitial pneumonia: imaging diagnosis, spectrum of abnormalities, and temporal progression. Proceedings of the ATS, January 1, 2006; 3(4): 307 - 314. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Perez IV, H. O. Coxson, J. C. Hogg, K. Gibson, P. F. Thompson, and R. M. Rogers Use of CT Morphometry To Detect Changes in Lung Weight and Gas Volume Chest, October 1, 2005; 128(4): 2471 - 2477. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. A. Lynch, J. D. Godwin, S. Safrin, K. M. Starko, P. Hormel, K. K. Brown, G. Raghu, T. E. King Jr., W. Z. Bradford, D. A. Schwartz, et al. High-Resolution Computed Tomography in Idiopathic Pulmonary Fibrosis: Diagnosis and Prognosis Am. J. Respir. Crit. Care Med., August 15, 2005; 172(4): 488 - 493. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. Azuma, Y. J. Li, S. Abe, J. Usuki, K. Matsuda, S. Henmi, Y. Miyauchi, K. Ueda, A. Izawa, S. Sone, et al. Interferon-{beta} Inhibits Bleomycin-Induced Lung Fibrosis by Decreasing Transforming Growth Factor-{beta} and Thrombospondin Am. J. Respir. Cell Mol. Biol., February 1, 2005; 32(2): 93 - 98. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. G. Parr, B. C. Stoel, J. Stolk, P. G. Nightingale, and R. A. Stockley Influence of Calibration on Densitometric Studies of Emphysema Progression Using Computed Tomography Am. J. Respir. Crit. Care Med., October 15, 2004; 170(8): 883 - 890. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. G. Parr, R. A. Stockley, A. Best, and D. Lynch Standardization of CT Densitometry [letter] * Drs Best and Lynch respond: Radiology, March 1, 2004; 230(3): 887 - 887. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
