Radiology
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH
QUICK SEARCH:   [advanced]


     

Published online before print December 19, 2006, 10.1148/radiol.2422060029

(Radiology 2006;242:563.)

A more recent version of this article appeared on December 1, 2006
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow All Versions of this Article:
2422060029v1
242/2/563    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Montaudon, M.
Right arrow Articles by Laurent, F.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Montaudon, M.
Right arrow Articles by Laurent, F.
© RSNA, 2006

Thoracic Imaging

Assessment of Airways with Three-dimensional Quantitative Thin-Section CT: In Vitro and in Vivo Validation1

Michel Montaudon, MD, PhD, Patrick Berger, MD, PhD, Gabriel de Dietrich, PhD, Achille Braquelaire, PhD, Roger Marthan, PhD, MD, José Manuel Tunon-de-Lara, MD, PhD and François Laurent, MD

1 From the Laboratory of Cellular Respiratory Physiology, Université Bordeaux 2, Bordeaux, France, and Institut National de la Santé et de la Recherche Médicale, E 356, F 33076, Bordeaux, France (M.M., P.B., R.M., J.M.T.d.L., F.L.); Department of Thoracic and Cardiovascular Imaging, CHU de Bordeaux, Hôpital du Haut-Lévêque, F 33604, Hôpital Cardiologique, avenue de Magellan, 33604 Pessac, France (M.M., F.L.); and Université Bordeaux 1, Talence, France (G.d.D., A.B.). Received January 6, 2006; revision requested March 7; revision received March 31; accepted May 3; final version accepted May 10. Supported by grants from Programme Hospitalier de Recherche Clinique received in 2002. Address correspondence to F.L. (e-mail: francois.laurent{at}chu-bordeaux.fr).


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCE IN KNOWLEDGE
 References
 
Purpose: To prospectively validate the ability of customized three-dimensional (3D) software to enable bronchial tree skeletonization, orthogonal reconstruction of the main bronchial axis, and measurement of cross-sectional wall area (WA) and lumen area (LA) of any visible bronchus on thin-section computed tomographic (CT) images.

Materials and Methods: Institutional review board approval and patient agreement and informed consent were obtained. Software was validated in a phantom that consisted of seven tubes and an excised human lung obtained and used according to institutional guidelines. In vivo validation was performed with multi–detector row CT in six healthy subjects (mean age, 47 years; range, 20–55 years). Intra- and interobserver agreement and reproducibility over time for bronchial tree skeletonization were evaluated with Bland-Altman analysis. Concordance in identifying bronchial generation was assessed with the {kappa} statistic. WA and LA obtained with the manual method were compared with WA and LA obtained with validated software by means of the Wilcoxon test and Bland-Altman analysis.

Results: WA and LA measurements in the phantom were reproducible over multiple sessions (P > .90) and were not significantly different from WA and LA assessed with the manual method (P > .62). WA and LA measurements in the excised lung and the subjects were not different from measurements obtained with the manual method (intraclass correlation coefficient > 0.99). All lobar bronchi and 80.8% of third generation bronchi, 72.5% of fourth generation bronchi, and 37.7% of fifth generation bronchi were identified in vivo. Intra- and interobserver agreement and reproducibility over time for airway skeletonization and concordance in identifying bronchial generation were good to excellent (intraclass correlation coefficient > 0.98, {kappa} > 0.54, respectively).

Conclusion: This method enables accurate and reproducible measurement of WA and LA on reformatted CT sections perpendicular to the main axis of bronchi visible on thin-section CT scans.

© RSNA, 2006


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCE IN KNOWLEDGE
 References
 
Thin-section computed tomography (CT) has been used to assess bronchial wall thickening in patients with airway diseases. However, visual assessment of bronchial wall thickening remains mainly subjective and poorly reproducible (1). Therefore, objective methods are needed to perform longitudinal studies and to compare airway dimensions before and after therapeutic intervention in patients with chronic diseases (2). Quantitative algorithms for analysis of CT images have already been designed and validated with phantom and animal studies in which airway dimensions were measured (3–6). However, limitations in these investigations include analysis of only one bronchus (4), manual delineation of bronchial contours (6), and restriction to bronchi running almost perpendicular to the transverse CT section (5). Because of the anatomy of the lung, most of the airways are likely to run obliquely rather than perpendicularly to the plane of the CT section. Therefore, for a two-dimensional thin-section CT image, there is an error in the method used to calculate the cross-sectional lumen area (LA) and wall area (WA).

The feasibility of designing a protocol to achieve three-dimensional (3D) quantification of bronchial parameters with multi–detector row CT and dedicated 3D software has been demonstrated (7,8). However, the ability of software to enable the identification of bronchial generations, visualization of true orthogonal sections of the main axis of bronchi, and accurate measurement of LA and WA has not been validated.

Thus, the purpose of our study was to prospectively validate the ability of customized 3D software to enable bronchial tree skeletonization, orthogonal reconstruction of the main bronchial axis, and measurement of cross-sectional WA and LA of any visible bronchus on thin-section CT images.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 ADVANCE IN KNOWLEDGE
 References
 
Institutional review board approval was obtained; all subjects agreed to participate in this study and provided informed consent. A normal lung specimen from the laboratory of anatomy at our institution (Institut National de la Santé et de la Recherche Médicale, E 356) was used with permission and in accordance with institutional guidelines.

Study Design
A dedicated software tool for analyzing bronchi in three dimensions was used to skeletonize the bronchial tree and reconstruct two-dimensional thin-section CT images orthogonal to the main bronchial axis of bronchi visible on multi–detector row CT scans in order to measure WA and LA. This software enabled analysis of the bronchial tree according to bronchial generation. Reproducibility of bronchial tree skeletonization was evaluated in vivo. The accuracy of measurements was assessed in vitro with a phantom made with silicone tubes and an excised human lung.

Image Analysis Software
We used software that was based on a framework used to segment tubular organs (9,10), implemented on a personal computer (Maxdata, Würselen, Germany), and run with a Linux operating system (Mandrake Linux, version 9.1; Mandriva, Moreno Valley, Calif). This software program followed two principal steps: First, it reformatted thin-section CT images perpendicular to the bronchial axis. Second, it measured WA and LA on these reformatted images (Fig 1).


Figure 1
View larger version (107K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1: A dedicated software tool that was used to analyze bronchi in three dimensions was applied to chest multi–detector row CT images. A, Volumetric data acquired with thoracic CT were used to reconstruct 1-mm-thick images (arrows indicate the direction of the propagation algorithm) from which, B, a propagation algorithm was used to obtain a binary volume based on bi-thresholding (frontal view). The area in which measurements were obtained is shown (arrowhead). The resulting image was automatically generated after the observer placed a seed point in the trachea. The airway skeleton was computed and superimposed on the binary volume. C, The resulting 3D image can be seen at various angles (oblique view). The accuracy of the central computation of the central axis can be checked to choose the most appropriate segment for orthogonal two-dimensional reformation. The area in which measurements were obtained is shown (arrowhead). D, A peripheral obliquely orientated bronchus (arrow) is shown on a native transverse thin-section CT scan. Reconstructions of cross-sectional 1-mm-thick CT scans of the selected bronchus perpendicular to the central axis were obtained. E, Three contiguous magnified thin-section CT scans are shown. F, The thin-section CT scan on which measurements were obtained was carefully selected by the observer. Scans that showed the least contiguity with surrounding vessels were chosen. G, A Laplacian of Gaussian algorithm was used to segment the designed airway and measure LA and WA.

 
All thin-section CT images obtained with volumetric acquisition were imported into the workstation in Digital Imaging and Communications in Medicine format by using a local area network (DXMM; Medasys, Gif-sur-Yvette, France). The first step in this procedure was to perform automatic segmentation of the bronchial lumen on the basis of noise filtering (11,12); this was followed by bi-thresholding (ie, thresholding between two specified values; in this case, –1024 HU and –920 HU) and morphologic erosion. An observer placed a seed point within the lumen of the trachea on the first CT section of the volume. Thereafter, the software automatically reconstructed the bronchial binary volume by using a volume-growing algorithm from the voxels previously bi-thresholded and connected to the seed point. The binary volume was considered complete when all the connected voxels were recruited. With this fully automatic procedure, the reconstructed volume could not be completed if there was not a connection between voxels.

A geodesic distance transformation method propagating from the selected seed point allowed extraction of isocontours on the binary volume surface. Automated computation of the local center of each isocontour was then performed, and local centers were connected to obtain a simplified image of the bronchial tree (ie, its skeleton). The computed skeleton could be superimposed on the binary volume; therefore, information about bronchial 3D orientation and division order could be obtained. This allowed the radiologist to verify the adequacy between the bronchial tree skeleton and the 3D binary volume. The spatial coordinates of the local centers, including those corresponding to bronchial bifurcations, could be extracted. The spatial coordinates were obtained by referring to the upper left voxel of the first section of the CT volume acquisition and could be converted to millimeters. The software allowed rotation and enlargement of the 3D reconstructed binary volume to determine the bronchial generation and spatial orientation. Finally, reformatted thin-section CT images orthogonal to the bronchial tree skeleton were reconstructed between two contiguous local centers to yield 1-mm-thick cross-sectional images of the selected bronchus perpendicular to its central axis at 1-mm intervals.

Successive magnified sections of the selected bronchus were visualized at the center of the reformatted images, and for each section selected for further detection of airway contours its position along the skeleton of the bronchial tree can be seen at the same time. The central computation of the selected airway and its surrounding vessels could be used for precise matching between two acquisitions. By rotating the skeleton, an observer could determine the generation of each bronchus after the successive bronchial divisions had been identified. Any missing bifurcation of a bronchus on the image of the bronchial tree skeleton could be detected retrospectively on the reconstructed two-dimensional sections since there was no gap between the reformatted cross sections. Therefore, if a bifurcation was missed at skeletonization but recognized when looking at reformatted sections, the order of the bronchus distal to the lacking bifurcation could then be corrected by the observer. Although it was possible to manually expand the number of bronchi taken into account for the binary volume construction, we kept the method fully automated for the purpose of our study.

From the reformatted sections, a Laplacian of Gaussian algorithm was used to segment the designed airway and measure LA and WA, as previously described and validated (5).

Validation of Skeletonization Reproducibility: An in Vivo Study
This study prospectively included six healthy volunteers (five men, one woman; mean age, 47 years; age range, 20–55 years) without a clinical history of pulmonary disease and with normal pulmonary function test results.

Volumetric acquisitions were performed with a 16-section multi–detector row CT unit (Somatom Sensation 16; Siemens, Erlangen, Germany). The following parameters were used: 120 kV, 53 mAs, 0.75-mm native section thickness, 1-mm reconstruction section thickness, 1-mm reconstruction interval, 284–380-mm2 field of view, and high-spatial-frequency algorithm. Two observers (M.M., F.L.) with 10 and 15 years, respectively, of experience in chest CT image interpretation performed intra- and interobserver reproduction of skeletonization and identification of bronchial generations; one observer (M.M.) repeated the analysis. Each observer noted the spatial coordinates of all local centers and was asked to attribute each bronchus to a specific generation according to the skeletonized image created with the method described in the Image Analysis Software section of this article. One observer (M.M.) repeated the entire procedure twice at 1-month intervals. To check intra- and interobserver reproducibility of skeletonization, spatial coordinates of each local center and of those corresponding to bronchial bifurcations extracted from the skeletonized image constructed by each observer were compared. In addition, the number of bronchi attributed to each generation by each observer was compared with a theoretical number based on the Boyden classification (13), in which segmental bronchi is considered the third generation.

CT images obtained at 1-year intervals could be compared in five of the six subjects. This allowed us to evaluate the ability of the software to assess follow-up measurements at the same bronchial level. One observer performed the entire procedure for both examinations of all five subjects. Since slightly different fields of view were used, spatial coordinates of bifurcations were converted to millimeters. Because the CT examinations were started at slightly different levels, every coordinate was corrected by using the difference between coordinate values of tracheal bifurcations in both examinations.

Accuracy of Measurements: An in Vitro Study
Phantom study.—To validate the measurement step, a phantom was made with seven silicone tubes (mean attenuation, 725.6 HU; range, 474–945 HU) embedded in a foam block (mean attenuation, –888.6 HU; range, –879.7 to –893.1 HU). Multi–detector row CT scans were acquired in the helical mode with the following parameters: 1-mm collimation, 120 kV, 58 mAs, and 420-msec rotation time. CT images were acquired at 15° intervals from 0° (strictly perpendicular to the long axis of the tubes) to 90° to measure tube sections at different oblique angles. CT data were reconstructed with a high-spatial-frequency algorithm, 1-mm section thickness, and 384-mm2 field of view; images were displayed on the monitor with parenchymal window width (1800 HU) and level (–600 HU) settings. Images were then transferred to the workstation, and WA and LA of each tube at each orientation were measured four times by one observer (M.M.) with the dedicated software tool to evaluate intraobserver variation. The actual (reference standard) LA and WA of the hollow silicone tubes were manually measured by an independent observer (F.L.) with commercial software (Scion Image, 4.0.2; Scion, Frederick, Md) (14) on high-resolution digital photographs (Fujifilm Finepix F810; Fuji Photograph Film, Tokyo, Japan) of the cross-section of each tube. The mean value of three consecutive measurements was calculated for each tube.

Excised lung study.—The excised and fixed right lung of a 60-year-old white woman was used to evaluate the accuracy of LA and WA measurements on a branched 3D structure. Helical multi–detector row CT images of the entire lung were acquired with the following parameters: 1-mm collimation, 120 kV, and 38 mAs. CT data were reconstructed with a high-spatial-frequency algorithm, 1-mm section thickness, and 384-mm2 field of view; CT images were displayed on the monitor with parenchymal window width (1800 HU) and level (–600 HU) settings. Images of 15 bronchi were then transferred to the personal computer and subsequently analyzed by one observer (M.M.) with dedicated software for 3D analysis of bronchi, or they were transferred to the CT workstation (Leonardo; Siemens) to obtain two-dimensional reformatted CT sections at the same anatomic level. LA and WA were measured on reformatted CT sections by another observer (F.L.) with the manual method.

Evaluation of the Entire Procedure: An in Vivo Study
The accuracy and reproducibility of WA and LA measurements have been previously evaluated on native transverse thin-section CT scans (5). To validate the entire procedure (ie, skeletonization followed by WA and LA measurement), 19 airways selected from CT scans obtained in one of the six subjects (randomly selected) were used. Selected airways ranged from the first generation (trachea) to the seventh generation. Since our goal was to evaluate the ability of the software to accurately measure the obliquely or horizontally oriented airways of various dimensions, 19 airways were selected (ie, trachea; right main bronchus; right and left B1, B2, and B3; left B7a; and 14 branches of the segment right S3). A helical scan of the entire lung was acquired with multi–detector row CT (0.75-mm collimation, 120 kV, and 38 mAs). CT data were reconstructed with a high-spatial-frequency algorithm, 1-mm section thickness, and 318-mm2 field of view. CT scans were displayed on the monitor with parenchymal window width (1800 HU) and level (–600 HU) settings. Images were then transferred either to the personal computer and analyzed with the dedicated software tool by one observer (M.M.) or to the CT workstation to obtain two-dimensional reformatted images at the same anatomic level. Because we needed to extrapolate the circumference of the bronchial cross section in contact with the pulmonary artery, we chose the section on which contact between the bronchus and pulmonary artery sections was minimal. LA and WA were measured on the latter section by an observer (F.L.) who used the manual method as the reference standard.

Statistical Analysis
Concordance between both observers in assigning each bronchus to a specific generation was assessed with the {kappa} statistic. Intra- and interobserver reproducibility of skeletonization, as well as reproducibility of two scans obtained in the same subject, was evaluated with Bland-Altman analysis performed with spatial coordinates of local centers (15–17).

For the seven tubes in the phantom, WA and LA measurements obtained at various angles were compared with WA and LA measurements obtained with the manual method by means of nonparametric analysis of variance (Kruskal-Wallis test). Accuracy of measurements obtained in the phantom was evaluated with the Kruskal-Wallis test. Accuracy of both in vitro (excised lung) and in vivo measurements was evaluated with a paired Wilcoxon test. Agreement between data obtained with dedicated software and data obtained manually was assessed with Bland-Altman analysis. Log transformations were performed with data that did not follow a normal distribution by using skewness, kurtosis, and omnibus tests. After log transformation, WA and LA followed a normal distribution, whereas spatial coordinates did not. However, according to Altman and Bland (18), distribution can be ignored for samples of hundreds of observations.

All analyses were performed with NCSS software (NCSS 2001; NCSS Statistical Software, Kaysville, Utah). A P value less than .05 was considered to indicate a statistically significant difference.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 ADVANCE IN KNOWLEDGE
 References
 
Skeletonization
With use of the fully automated skeletonization procedure, the number of reconstructed bronchi per generation compared with the theoretical number of reconstructed bronchi decreased dramatically beyond the fifth generation (Table 1). {kappa} Values for intra- and interobserver concordance of the ability to assign a generation to a visible bronchus on the skeleton were excellent ({kappa} ≥ 0.95), and concordance over time was good ({kappa} ≥ 0.54) (Table 1).


View this table:
[in this window]
[in a new window]

  		Table 1. Concordance of the Bronchial Generation Assessment

 
Bland-Altman analysis was used to compare coordinates from a mean of 344 local centers (range, 214–505 centers), including a mean of 47.6 bronchial divisions (range, 30–80 divisions) per subject extracted by both observers. Intra- and interobserver agreement and correlation coefficients over time were high (Table 2, Fig 2). Lack of intra- and interobserver agreement was low (Table 2). Considering comparison over time, the lack of agreement (bias estimated with the mean difference and the standard deviation of the difference) was slightly greater, particularly for the z coordinate (Table 2, Fig 2). The error measurement was assessed with the within-subject standard deviation and was found to be minimal (Table 2).


View this table:
[in this window]
[in a new window]

  		Table 2. Agreement between Spatial Coordinates Defining Skeletonization of Airway Trees

 

Figure 2
View larger version (13K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2a: (a) Graphs of data from correlation over time for x, y, and z spatial coordinates (left, middle, and right graphs, respectively) of each bronchial bifurcation. (b) Graphs of means of measurements over time for x, y, and z spatial coordinates (left, middle, and right graphs, respectively) are plotted against their difference according to Bland-Altman analysis. Solid lines correspond to the mean difference. Dashed lines correspond to the mean difference ± 2 standard deviations and the 95% confidence interval. Lack of agreement was greater for z coordinates. ICC = intraclass correlation coefficient, Obs1 = observer 1.

 

Figure 2
View larger version (19K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2b: (a) Graphs of data from correlation over time for x, y, and z spatial coordinates (left, middle, and right graphs, respectively) of each bronchial bifurcation. (b) Graphs of means of measurements over time for x, y, and z spatial coordinates (left, middle, and right graphs, respectively) are plotted against their difference according to Bland-Altman analysis. Solid lines correspond to the mean difference. Dashed lines correspond to the mean difference ± 2 standard deviations and the 95% confidence interval. Lack of agreement was greater for z coordinates. ICC = intraclass correlation coefficient, Obs1 = observer 1.

 
Accuracy of Measurements
Phantom study.—Mean values of WA (range, 5.5–92.8 mm2) and LA (range, 1.1–43.3 mm2) in the seven silicon tubes assessed with dedicated software and the manual method are shown in Table 3. Comparison of WA and LA measurements obtained on thin-section CT scans at various angles with the manual method and with dedicated software did not show any significant difference (P = .615 and P = .995, respectively; Kruskal-Wallis test). WA and LA measurements obtained with dedicated software were reproducible over multiple sessions (P = .991 and P = .989, respectively; Kruskal-Wallis test).


View this table:
[in this window]
[in a new window]

  		Table 3. Comparison of Measurements Obtained with Dedicated Software and Those Obtained with the Manual Method at Various Angles on Thin-Section CT Images of the Phantom

 
Excised lung study.—Fifteen bronchi with a mean WA of 18.1 mm2 (range, 6.6–47.3 mm2) and a mean LA of 11.8 mm2 (range, 1.5–60.0 mm2) were evaluated on thin-section CT scans with the dedicated software tool and the manual method. WA and LA were measured with the dedicated software tool; these measurements were not significantly different from reference values (P > .33, Wilcoxon test; Table 4). For both WA and LA, strong correlation and low lack of agreement were found for both methods (Table 4). The error measurement, which was assessed with the within-subject standard deviation (0.03 for WA, 0.02 for LA), was low. Standard deviations of WA and LA were assessed with dedicated software and were shown to not correlate with the mean values of WA and LA, respectively (Table 4).


View this table:
[in this window]
[in a new window]

  		Table 4. Excised Lung and in Vivo Study: Reliability of WA and LA Measurements

 
In vivo study.—The 19 airways were evaluated both (a) with reconstructed images created with dedicated software and (b) with multiplanar reformations (Fig 3) and had a WA that ranged from 6.45 to 97.9 mm2 (mean, 25.6 mm2) and an LA that ranged from 2.23 to 312.8 mm2 (mean, 37.5 mm2). WA and LA measurements obtained with dedicated software were not significantly different from reference values (P > .44, Wilcoxon test; Table 4). For both WA and LA, strong correlation and low lack of agreement were found for both methods (Table 4, Fig 4). The error measurement, which was assessed with the within-subject standard deviation, was low for both WA and LA (0.01 and 0.02, respectively). Standard deviations of WA and LA obtained with software did not correlate with mean values of WA and LA (Table 4, Fig 4).


Figure 3
View larger version (114K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3: Images obtained during the in vivo examination. A, Native transverse thin-section CT scan shows the selected bronchus is B2a (arrow). B, Binary volume and skeleton of right S2 shows B2a (arrowhead) and the location in which reformatted thin-section CT scans were obtained. C, Multiplanar reformatted thin-section CT scan perpendicular to the main bronchial axis. Outlined area indicates the area shown in E. D, Corresponding CT scan reconstructed with dedicated software. Outlined area indicates the area shown in F. E, Magnified image of C shows internal and external contours of the bronchus as assessed by an observer using the manual method. (Original magnification, x3.) F, Magnified image of D with automatic detection of internal and external airway contours with dedicated software. (Original magnification, x3.)

 

Figure 4
View larger version (24K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 4a: Graphs of data collected during the in vivo examination. x = WA, {diamondsuit} = LA. (a) WA and LA measurements obtained with dedicated software were plotted against measurements obtained manually. The diagonal line corresponds to the line of equality. There was a strong correlation between data obtained with software and data obtained manually, as assessed with the intraclass correlation coefficient (ICC). r1 = Pearson correlation coefficient. (b) Means of measurements are plotted against their difference according to Bland-Altman analysis. The solid line corresponds to the mean difference. The dashed lines correspond to the 2 standard deviations. For each standard deviation, a 95% confidence interval that corresponds to the irregular lines can be calculated. (c) Means of measurements are plotted against their standard deviations. r2 = Pearson correlation coefficient.

 

Figure 4
View larger version (26K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 4b: Graphs of data collected during the in vivo examination. x = WA, {diamondsuit} = LA. (a) WA and LA measurements obtained with dedicated software were plotted against measurements obtained manually. The diagonal line corresponds to the line of equality. There was a strong correlation between data obtained with software and data obtained manually, as assessed with the intraclass correlation coefficient (ICC). r1 = Pearson correlation coefficient. (b) Means of measurements are plotted against their difference according to Bland-Altman analysis. The solid line corresponds to the mean difference. The dashed lines correspond to the 2 standard deviations. For each standard deviation, a 95% confidence interval that corresponds to the irregular lines can be calculated. (c) Means of measurements are plotted against their standard deviations. r2 = Pearson correlation coefficient.

 

Figure 4
View larger version (26K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 4c: Graphs of data collected during the in vivo examination. x = WA, {diamondsuit} = LA. (a) WA and LA measurements obtained with dedicated software were plotted against measurements obtained manually. The diagonal line corresponds to the line of equality. There was a strong correlation between data obtained with software and data obtained manually, as assessed with the intraclass correlation coefficient (ICC). r1 = Pearson correlation coefficient. (b) Means of measurements are plotted against their difference according to Bland-Altman analysis. The solid line corresponds to the mean difference. The dashed lines correspond to the 2 standard deviations. For each standard deviation, a 95% confidence interval that corresponds to the irregular lines can be calculated. (c) Means of measurements are plotted against their standard deviations. r2 = Pearson correlation coefficient.

 

   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
ADVANCE IN KNOWLEDGE
 References
 
Our study results demonstrate that dedicated software allows an observer to determine bronchial generation and measure WA and LA on CT images reconstructed perpendicular to the main bronchial axis.

Since obliquity affects airway measurements, software that allows observers to determine the location of the central axis of the bronchus and the thin section perpendicular to this axis is needed. To address this issue, Wood et al (19) developed a threshold method to define airway LA leading to the lumen centerline to define the central axis of the airway so the angle of orientation could be measured. However, this method was developed before the advent of multi–detector row CT; therefore, it was accurate only for measuring structures larger than 2 mm in diameter (19). Nevertheless, other authors (20) have developed algorithms that correct the effect of airway orientations on LA and WA. Recently, 3D methods that allow skeletonization of the bronchial tree have been proposed (7,8). In our study, we used skeletonization of the bronchial tree to obtain sections orthogonal to the main axis of the bronchi. We have validated the reproducibility of both the skeletonization technique and the entire procedure. Ideally, such software should be fully automated. However, because of interruption of the bronchial wall secondary to low signal-to-noise ratio of the most distal bronchi, the fully automated procedure resulted in a dramatically decreased number of assessable bronchi beyond the fifth generation. In addition, a report (21) has shown that reproducibility of LA measurements is acceptable for bronchi with an internal diameter of more than 1 mm (LA = 0.79 mm2). Our analysis overcame the major limitation in the use of CT in quantitative analysis (ie, accurate measurement of LA and WA restricted to airways oriented approximately perpendicular to the imaging plane). Thus, unlike in previous studies that addressed bronchial measurement, the entire bronchial tree could be included in the analysis.

To circumvent interobserver variability and parallax errors ascribed to manual methods on the reformations orthogonal to the main bronchial axis, we used dedicated and validated software to detect internal and external contours of bronchi (5). This method is semiautomated and allows correction of the airway external wall contour despite its connection to a vessel, without assuming a perfect roundness and symmetry of the WA (5). Alternative methods have been proposed and used to determine airway contours. Manual tracing and measurement on the monitor screen with adjustment of window and level settings constitute the simplest method (6,22–24). More sophisticated automated and semiautomated methods involving the use of full width half maximum (25,26), thresholding and region growing (20,27,28), or mathematical morphology (7,29) have also been proposed.

Additional technical issues must be addressed. Irradiation burden must be kept to a minimum; thus, we used a low-dose multi–detector row CT protocol. However, low signal-to-noise ratio may have limited our study by impairing bronchial segmentation and the resulting fully automated skeleton. In addition, decreased image quality of paracardiac and middle lobe territories is a known drawback of multi–detector row CT. Electrocardiographic gating could have improved automated postprocessing, but it would have increased the radiation dose. Finally, the known relationship between airway caliber and lung volume leads to use of spirometric gating (30). To keep multi–detector row CT within the standards of a routine imaging examination, we did not use respiratory gating in this study. However, all examinations were performed in subjects at full inspiration, as this respiratory status has less of an effect on airway dimension variability (31).

In this study, we used geodesic front propagation to create the skeletonized image. This is a central axis computation technique that has been known to fail in segments of large caliber; however, this technique is robust when applied to small segments. The radiologist was always able to verify the appropriateness of the central axis computation on images that superimposed the bronchial tree skeleton on 3D reconstructed images of the bronchial tree. Another limitation of our study is the fact that although airway measurements were highly correlated with values obtained with the manual method, the accuracy of manual measurement cannot serve as a reference standard; therefore, some differences may remain between our results and the absolute values of airway cross sections in living subjects. Finally, when the bronchial cross sections show a large area of contact between the bronchus and the pulmonary artery, there remains a certain degree of inaccuracy that largely depends on the amount of contact. Thus, a manual editing technique or a semiautomatic segmentation procedure is needed for extrapolation of the outer contour from the rest of the bronchial circumference in contact with lung parenchyma.

We have validated an algorithm that can be used not only to measure cross-sectional WA and LA on CT images automatically reconstructed perpendicular to the main bronchial axis but also to identify the bronchial generation. This method has been applied in patients with cystic fibrosis (32).


   ADVANCE IN KNOWLEDGE
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCE IN KNOWLEDGE
References
 


   FOOTNOTES
 

Abbreviations: LA = lumen area • 3D = three dimensional • WA = wall area

Authors stated no financial relationship to disclose.

Author contributions: Guarantors of integrity of entire study, P.B., F.L.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, M.M., P.B., A.B., F.L.; clinical studies, M.M., G.d.D., F.L.; experimental studies, M.M.; statistical analysis, P.B., R.M.; and manuscript editing, M.M., P.B., R.M., J.M.T.d.L., F.L.


   References
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCE IN KNOWLEDGE
 References
 

  1. Park JW, Hong YK, Kim CW, Kim DK, Choe KO, Hong CS. High-resolution computed tomography in patients with bronchial asthma: correlation with clinical features, pulmonary functions and bronchial hyperresponsiveness. J Investig Allergol Clin Immunol 1997;7:186–192.[Medline]
  2. Bankier AA, Fleischmann D, De Maertelaer V, et al. Subjective differentiation of normal and pathological bronchi on thin-section CT: impact of observer training. Eur Respir J 1999;13:781–786.[Abstract]
  3. Little SA, Sproule MW, Cowan MD, et al. High resolution computed tomographic assessment of airway wall thickness in chronic asthma: reproducibility and relationship with lung function and severity. Thorax 2002;57:247–253.[Abstract/Free Full Text]
  4. Nakano Y, Muro S, Sakai H, et al. Computed tomographic measurements of airway dimensions and emphysema in smokers: correlation with lung function. Am J Respir Crit Care Med 2000;162:1102–1108.[Abstract/Free Full Text]
  5. Berger P, Perot V, Desbarats P, Tunon-de-Lara JM, Marthan R, Laurent F. Airway wall thickness in cigarette smokers: quantitative thin-section CT assessment. Radiology 2005;235:1055–1064.[Abstract/Free Full Text]
  6. Orlandi I, Moroni C, Camiciottoli G, et al. Chronic obstructive pulmonary disease: thin-section CT measurement of airway wall thickness and lung attenuation. Radiology 2005;234:604–610.[Abstract/Free Full Text]
  7. Fetita CI, Preteux F, Beigelman-Aubry C, Grenier P. Pulmonary airways: 3D reconstruction from multi-slice CT and clinical investigation. IEEE Trans Med Imaging 2004;23:1353–1364.[CrossRef][Medline]
  8. Palagyi K, Tschirren J, Hoffman EA, Sonka M. Quantitative analysis of pulmonary airway tree structures. Comput Biol Med 2006;36:974–996.[CrossRef][Medline]
  9. De Dietrich G. A modular algorithm for automatic slice positioning in tubular organs. In: Medical imaging and augmented reality. Hong Kong, China: IEEE, 2001; 163–167.
  10. de Dietrich G, Braquelaire A. A framework for tubular organs segmentation. In: Winter School of Computer Graphics. Plzen, Czech Republic: Union Agency—Science Press, 2004; 41–44.
  11. Lee JS. Digital image smoothing and the sigma filter. Comput Vis Graph Image Proc 1983;24:255–269.[CrossRef]
  12. Böhm D, Krass S, Kriete A, et al. Segmentbestimmung im computertomogramm der lunge: in-vitro validierung. In: Horsch A, Lehmann T, eds. Bildverarbeitung für die medizin 2000. Berlin: Springer, 2000; 168–172.
  13. Boyden EA. Segmental anatomy of the lungs: a study of the patterns of the segmental bronchi and related pulmonary vessels. New York, NY: McGraw-Hill, 1955.
  14. Kuszak JR, Al-Ghoul KJ. A quantitative analysis of sutural contributions to variability in back vertex distance and transmittance in rabbit lenses as a function of development, growth, and age. Optom Vis Sci 2002;79:193–204.[Medline]
  15. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:307–310.[CrossRef][Medline]
  16. Bland JM, Altman DG. Statistics notes: measurement error and correlation coefficients. BMJ 1996;313:41–42.[Free Full Text]
  17. Bland JM, Altman DG. Statistics notes: measurement error. BMJ 1996;313:744.[Free Full Text]
  18. Altman DG, Bland JM. Statistics notes: the normal distribution. BMJ 1995;310:298.[Free Full Text]
  19. Wood SA, Zerhouni EA, Hoford JD, Hoffman EA, Mitzner W. Measurement of three-dimensional lung tree structures by using computed tomography. J Appl Physiol 1995;79:1687–1697.[Abstract/Free Full Text]
  20. King GG, Muller NL, Whittall KP, Xiang QS, Pare PD. An analysis algorithm for measuring airway lumen and wall areas from high-resolution computed tomographic data. Am J Respir Crit Care Med 2000;161:574–580.[Abstract/Free Full Text]
  21. King GG, Carroll JD, Muller NL, et al. Heterogeneity of narrowing in normal and asthmatic airways measured by HRCT. Eur Respir J 2004;24:211–218.[Abstract/Free Full Text]
  22. Awadh N, Muller NL, Park CS, Abboud RT, FitzGerald JM. Airway wall thickness in patients with near fatal asthma and control groups: assessment with high resolution computed tomographic scanning. Thorax 1998;53:248–253.[Abstract/Free Full Text]
  23. Brown RH, Herold CJ, Hirshman CA, Zerhouni EA, Mitzner W. In vivo measurements of airway reactivity using high-resolution computed tomography. Am Rev Respir Dis 1991;144:208–212.[Medline]
  24. Okazawa M, Muller N, McNamara AE, Child S, Verburgt L, Pare PD. Human airway narrowing measured using high resolution computed tomography. Am J Respir Crit Care Med 1996;154:1557–1562.[Abstract]
  25. Amirav I, Kramer SS, Grunstein MM, Hoffman EA. Assessment of methacholine-induced airway constriction by ultrafast high-resolution computed tomography. J Appl Physiol 1993;75:2239–2250.[Abstract/Free Full Text]
  26. Brown RH, Mitzner W. The myth of maximal airway responsiveness in vivo. J Appl Physiol 1998;85:2012–2017.[Abstract/Free Full Text]
  27. McNitt-Gray MF, Goldin JG, Johnson TD, Tashkin DP, Aberle DR. Development and testing of image-processing methods for the quantitative assessment of airway hyperresponsiveness from high-resolution CT images. J Comput Assist Tomogr 1997;21:939–947.[CrossRef][Medline]
  28. Swift RD, Kiraly AP, Sherbondy AJ, et al. Automatic axis generation for virtual bronchoscopic assessment of major airway obstructions. Comput Med Imaging Graph 2002;26:103–118.[CrossRef][Medline]
  29. Prêteux F, Fetita C, Capderou A, Grenier P. Modeling, segmentation and caliber estimation of bronchi in high resolution computerized tomography. J Electron Imaging 1999;8:36–45.
  30. Brown RH, Mitzner W. Effect of lung inflation and airway muscle tone on airway diameter in vivo. J Appl Physiol 1996;80:1581–1588.[Abstract/Free Full Text]
  31. Becker MD, Berkmen YM, Austin JH, et al. Lung volumes before and after lung volume reduction surgery: quantitative CT analysis. Am J Respir Crit Care Med 1998;157:1593–1599.
  32. Montaudon M, Berger P, Sacher A, et al. Bronchial measurement with three-dimensional quantitative thin-section CT in patients with cystic fibrosis. Radiology 2006;242:573–581.



This article has been cited by other articles:


Home page
 RadiologyHome page
M. Montaudon, P. Berger, A. Cangini-Sacher, G. de Dietrich, J. M. Tunon-de-Lara, R. Marthan, and F. Laurent
Bronchial Measurement with Three-dimensional Quantitative Thin-Section CT in Patients with Cystic Fibrosis
Radiology, December 1, 2006; 242(2): 573 - 581.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow All Versions of this Article:
2422060029v1
242/2/563    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Montaudon, M.
Right arrow Articles by Laurent, F.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Montaudon, M.
Right arrow Articles by Laurent, F.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH
RADIOLOGY RADIOGRAPHICS RSNA JOURNALS ONLINE