INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

High resolution computed tomograpy (HRCT) is being used with increasing frequency as a research tool for studying airway dimensions in obstructive lung disease (1). The smallest voxel size that can be achieved with HRCT is 0.25 × 0.25 × 1 mm = 0.0624 mm³, when a slice thickness of 1 mm is used in conjunction with the smallest field of view (FOV) (13-cm diameter) and a matrix size of 512 × 512. With use of these parameters, HRCT can theoretically resolve the dimensions of airways approximately 1 to 2 mm in diameter under ideal conditions (i.e., when airways are oriented perpendicularly to the plane of imaging, and when there is minimal artifact from respiratory or cardiac motion).

Digital HRCT image data lend themselves to computerized manipulation, such as with different reconstruction algorithms (e.g., "bone algorithm"), and to magnification of specific regions of interest to suit a variety of applications. Despite this enormous potential, assessment of HRCT for clinical purposes is subjective. To use the full potential of HRCT to characterize airway dimensions and to evaluate acute changes, such as agonist-induced airway narrowing, validated, automated, and nonsubjective image analysis techniques are required. Four quantitative algorithms for analysis of HRCT images have been validated through phantom studies. McNamara and colleagues modified a method developed by Webb and associates (8), based on visual analysis of photographed images. They found that it was crucial to use a window level of 450 HU (9). Amirav and coworkers developed a computerized algorithm for measuring airway lumen area (Ai) that has the advantages of less subjectivity and greater speed than the method of McNamara and colleagues (10). Both of these algorithms have been used to quantify the magnitude and distribution of airway narrowing in excised animal lungs (9) and in animal lungs in vivo (11, 12), as well as in normal and asthmatic subjects (13). Wood and colleagues (14) developed an algorithm to measure Ai and the airway angle of orientation using a three-dimensional reconstruction of the lung. This technique allowed cross-sectional images of the airway to be generated irrespective of airway orientation. All of the algorithms described here have been developed primarily to examine airway lumen dimensions. The accuracy of measurements of airway wall area (Awa), and the effects of angled airway orientation on measurements of Awa made with these algorithms, have not been previously assessed.

The aim of the present study was to develop a computerized algorithm (hereafter referred to as computed tomographic airway morphometry [CTAM]) to measure Ai and Awa from HRCT images, using an airway phantom made of artificial materials and the airways in excised lung as "gold standards." We also compared the results for airway dimensions measured using CTAM with those obtained by the visual image analysis method (9) (hereafter referred to as the manual method). Since most airways in vivo do not run in the craniocaudal direction, but are oriented at various angles to this plane, we also included a measure of the airways' angle of orientation in CTAM.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Study Design

Two studies were performed. In the first, we used a phantom made of plastic tubes to evaluate the accuracy of the CTAM in the measurement of Ai, Awa, and the angle of orientation of the tubes. Measurements were also made with the manual method (9). In the second study we used two excised pig lungs, which were fixed by exposure to formalin fumes at constant pressure so that the lung structures retained similar densities to their in vivo states and remained filled with air. The fixed lungs were then cut into approximately 1-cm-thick slices. The cut surfaces were scanned with HRCT, and the airway dimensions measured from HRCT images, using both CTAM and the manual method, were then compared with measurements of airway dimensions made with direct planimetry of the cut surface.

Artificial Phantom Study

A lung phantom was made by embedding plastic syringe barrels into cylindrical holes made in a polystyrene foam block. The syringes were cut in cross section and their dimensions were measured with an optical micrometer to the closest 0.05 mm (Table 1). Holes corresponding to the outer diameter of each syringe barrel were bored perpendicular to the face of the polystyrene foam block, and the syringe barrels were inserted into the holes. HRCT scans were obtained with the barrels angled at 0°, 10°, 20°, 30°, 40°, and 50°, by referencing the front of the block to the laser guide of the CT scanner. A correction was made on all Ai and Awa measurements for the angle of orientation (corrected Ai = measured Ai/cos ), which allowed us to estimate areas as if the airways were imaged in cross-section. Scans were done with a CT9800 whole-body scanner (General Electric, Milwaukee, WI) in helical mode, using 1-mm collimation, a pitch of 1, a 20-cm FOV, a 512 × 512 matrix, a 120 kV peak, and 200 mA current. The CT image data were reconstructed with a high spatial frequency algorithm (bone algorithm; General Electric).

Excised Pig Lung

The lungs of a 27-kg and a 45-kg pig were removed after ligation of the pulmonary vessels. The lungs were fixed in inflation with formalin fumes, using the method described by Weibel and Vidone (15). Care was taken not to puncture the pleural surface during removal of the lung, and any small tears were oversewn. The lungs were initially fully inflated by applying a transpulmonary pressure of 30 cm H₂O in a sealed glass container (Figure 1). They were then maintained in inflation, using a surrounding pressure of 25 cm H₂O. Formalin steam, generated by boiling a diluted formalin solution (2 parts 37% formalin diluted with 1 part water), was passed into the airways during lung inflation at a pressure of 5 cm H₂O. Steam can pass into the lungs under these conditions because of leaks through microscopic tears in the pleura. The steaming process was continued for 2 h, during which drying and formalin absorption fixed the lungs, which remained inflated after removal of the pressure source. The lungs weighed 273 g and 636 g immediately before inflation, and 293 g and 702 g, respectively, immediately after fixation. The lung surface was then coated with gelatin to prevent drying. Once the gelatin was dry, the lung was cut into transverse sections from 1.5 to 2 cm thick, which were stored in plastic wrap to prevent drying.

View larger version (15K): [in this window] [in a new window]   Figure 1.   Schematic representation of the apparatus used for inflation-fixation of excised pig lung with formalin steam. Thirty percent formalin was boiled and the vapor was passed into the trachea under approximately 5 cm H2O pressure. The lung was inflated to 25 cm H2O negative pressure, as measured with a water manometer. The lung turned brown when it was adequately fixed.

Airways on the cut surface were visualized under a dissecting microscope, and the images were digitized with a high-resolution video camera attached to the dissecting microscope. One-millimeter calibration markers were included in the images (Figure 2). Images were transferred to a personal computer (PC) for quantitative analysis. Ai and the total area within the outer airway border (Ao) were measured from conventional graphic computations, using an in-house software package. The perimeters of the airway lumen and outer airway border were determined by eye and were traced with the computer mouse cursor. Ao, Ai, and lumen perimeter were measured from the tracings. Because of variable magnification, the area represented by each pixel was determined for each image from the calibration markers, in order to calculate the enclosed areas. A single observer (G.G.K.) made three repeated measurements of both Ai and Ao, and the means were used as the gold standards for each of these parameters. Awa was calculated as Ao  Ai.

View larger version (33K): [in this window] [in a new window]   Figure 2.   Digitized image of the cut surface of a lung slice (A) allows airways to be matched with those in the HRCT image (C ). Another digitized image, made through a dissecting microscope (B), is used for measuring Ai and Awa.

During HRCT scanning, the lung slices were held in position by placing the cut surface against a cardboard sheet, which was oriented in the scanning plane by using the laser guide of the CT scanner. The first HRCT slice was at the cut surface of the lung. Because the cut surface was not perfectly flat, the next contiguous 1-mm slice was used for analysis if parts of the cut surface were incomplete in the first slice. The excised pig lungs were scanned with the same scanner, settings, and reconstruction algorithm that were used for the artificial phantom.

Image Analysis: CTAM

The CT image data were transferred to a UNIX-based work station (SGI Indy; Silicon Graphics, Mountainview, CA) to measure Ai and Ao using CTAM. CT image data consisted of a 512 × 512 matrix of pixels with unique CT numbers, which represented the X-ray attenuation values of the (512 × 512 =) 262,144 pixels in each CT slice. HU were related to density according to the following formula:

(1)

The algorithm was written to run in the PV WAVE software package (Visual Numerics Inc., Boulder, CO).

Figures 3A and 3B show the steps in the image analysis algorithm for measuring Ai. The airway selected for measurement is identified, and its approximate center is marked by using the mouse cursor; magnification of the airway is then centered on this point (Figure 3A). The algorithm defines the pixel marked as the center of the lumen as the first Ai pixel. The pixels surrounding this "seed" pixel are then examined and are also labeled as Ai if their values are less than the threshold value (Figure 3B). Ai is calculated by multiplying the number of pixels by the pixel area (0.15 mm²). Ai was measured across a range of threshold values to determine the value that provided the most accurate measurements.

View larger version (123K): [in this window] [in a new window]   Figure 3.   Screen images seen during airway analysis with the CTAM procedure. (A) The airway to be measured is identified, the image is enlarged and the approximate center of the airway is identified with the PC cursor. (B) All adjoining pixels that have values less than the threshold value are identified, thereby defining Ai. (C ) A circle is drawn around the airway. (D) Iterative erosions of the circle, based on a score, continue until the outer border of the airway is identified. All pixels within that border are then highlighted. (E ) Two further erosions are completed before Ai pixels are subtracted to identify the airway wall seen in (F ).

A separate module of the algorithm also allows determination of the angle of the airway, relative to the long axis of the scanner. The center of the airway lumen is determined from the centroid. The x-y displacement between centroids of adjacent slices is measured, and the airway angle can be calculated from the slice thickness. The weighted mean angle of the airway over any number of slices can be calculated. The angles in the artificial phantom were measured over three contiguous slices: the slice from which the Ai and Awa measurements were made, and the slices on either side of it. The airway angles for the excised lung were not measured, since the gold standard measurements from the cut surface represented oblique sectioning, the angle of orientation being identical for planimetry and HRCT scanning.

Awa was more difficult to determine because variations in airway wall appearance and proximity to vessels meant that traditional threshold segmentation often produced a discontinuous airway wall. We therefore used score-guided erosion (SGE) to estimate Awa (16). SGE is an edge-finding algorithm, and incorporates knowledge that the airways are roughly circular and of high density relative to lung parenchyma. A "score map" was created from the image data by using a commonly employed edge-detection technique. A copy of the image was moved by one pixel in the x and y directions, so that it was out of phase compared with the original image. The values of all pixels in one image were then subtracted from the values resulting from the corresponding pixels in the other image. The absolute values resulting from the subtraction of each pixel from its corresponding pixel then formed the score map. This meant that pixels in the score map had higher values at locations where an edge was present (i.e., the more obvious the edge, the bigger the difference in attenuation values of adjacent pixels at the edge, and the higher the score value of the pixels at the edge).

An initial annular mask was drawn, centered on the airway (Figure 3C). The mask was iteratively eroded to maximize the score while preserving the connected-ring topology of the airway wall (Figure 3D). The algorithm allowed manual truncation from Ao of pixels that appeared to represent blood vessel rather than airway wall as judged by the operator's eye, on the basis of the assumption that the airway was of the same thickness where it contacted the vessel as it was elsewhere. After truncation, the algorithm also allowed further erosion of the outer rim of pixels to symmetrically reduce the area of Ao initially measured with the SGE technique. For measurements of Ao in the phantom, no additional erosions were made (Figure 3E). Ai was subtracted from Ao to give Awa (Figure 3F). All data presented for Awa were generated by using these procedures.

Image Analysis: Manual Method

HRCT images of the artificial phantom were photographed onto X-ray film at a window level of 450 HU and at window widths of 500 HU and 1,400 HU, and HRCT images of the pig lung phantom were photographed at a window level of 450 HU and a window width of 1,000 HU. The images were scanned and enlarged with a flatbed scanner, and were transferred to a PC for quantitative measurments of Ai and Ao, using the Scion Image software package (Scion Corportion, Frederick, MD). The airway lumen and outer wall perimeters were traced, and the areas contained within the perimeters were calculated, using the CT calibration scale routinely provided on all HRCT films. The means of triplicate measurements were used in the data analyses.

Data Analyses

Data are presented as mean ± SEM. Differences between HRCT airway measurements and actual dimensions were examined using the methods described by Bland and Altman (17). The accuracy of the HRCT measurements made with the CTAM algorithm were expressed as the 95% confidence interval (CI) of the estimate t_0.05 × SD_differences/, where t_0.05 is the critical t value for a two-sided test at the 0.05 level and SD_differences is the standard deviation of the differences between actual phantom and measured airway dimensions. Relationships between parameters were examined by using Pearson's correlation coefficient.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Phantom Study: CTAM

The densities of the phantom airway wall and "parenchyma" measured with HRCT were 0.79 g/ml and 0.05 g/ml, respectively. The mean errors for the measurement of Ai for all airways were determined for CT-number thresholds of 300, 350, 400, 450, and 500. These mean errors were plotted against the CT-number threshold, and the equation of the line of best fit was found to be: mean error (mm²) = 0.03 × threshold + 12.6. A CT-number threshold of 447 g/ml (577 HU) corresponded to a mean error of zero, and a threshold of 450 g/ml (574 HU) was therefore used in all measurements of Ai of phantom airways. The angle of orientation was measured for each airway, and the measured Ai and Awa were corrected for this angle. The mean error for angle measurement was 0.5 ± 0.4°. Figures 4A and 4B show the absolute errors of measurement of Ai and Awa for individual tubes of varying size when imaged at increasing angles. Figures 4C and 4D show the errors as a percentage of the true areas. The data show that Ai was accurately measured and that the errors in its measurment were unrelated either to angle of orientation or to airway size. The errors for the measurement of Awa were larger than those for Ai, and were dependent on airway size, being greater the larger the airway, but again were unrelated to the angle of orientation. The relative errors of Ai and Awa measurements (Figures 4C and 4D) were greatest for the smallest tube ("airway"), but were relatively constant for the larger tubes.

View larger version (24K): [in this window] [in a new window]   Figure 4.   Absolute (A and B) and percent (C and D) errors of measurement of Ai and Awa in an artificial phantom, using CTAM. The x-axis shows the true area measured from the tubes cut in cross-section. The y-axis shows the difference between true area and the area measured from HRCT in which a correction was made for the airway angle.

Phantom Study: Manual Method

Figures 5A and 5B show the absolute errors for the measurement of Ai and Awa, respectively, from CT images photographed at a window level of 450 HU and window width of 1,400 HU. Figures 5C and 5D show the errors relative to the true values of Ai and Awa. The Ai and Awa measurements were corrected for the known angle of orientation, even though it is not possible to measure the angle with the manual method in order to perform a post hoc correction. However, this procedure allowed errors to be directly compared with those of the CTAM method. Ai was underestimated, whereas Awa was overestimated. The overestimation of Awa was clearly dependent on both airway size and airway angle (Figures 5A and 5B), and the errors in measuring Ai and Awa were larger than the corresponding measurement errors for CTAM. The percentage errors were again largest for the smallest tube (airway), and were relatively constant for the larger airways (Figures 5C and 5D).

View larger version (25K): [in this window] [in a new window]   Figure 5.   Absolute (A and B) and percent (C and D) errors of measurement of Ai and Awa in an artificial phantom, using the manual method. The x-axis shows the true area measured from the tubes cut in cross-section. The y-axis shows the difference between true area and the area measured from HRCT in which a correction was made for the known angle (since it is not possible to measure the angle with this method). Note the different scales of the y-axes in the graphs shown in A through D.

Volume averaging will also yield a wider range of attenuation values for the airway wall. In the presence of a wide range of values, the window widths could then affect the measurement of Awa. The mean Awa of all airways at all angles was 45.5 ± 4.6 mm² when measured from images photographed at window widths of 500 HU and a window level of 450 HU; this was less than the mean of 52.6 ± 5.6 mm² measured from images photographed at 1,400 HU and a window level of 450 HU (p < 0.01) (Figure 6).

View larger version (23K): [in this window] [in a new window]   Figure 6.   Awa measurements made from an artificial phantom with the manual method, from images photographed at a window level of 450 HU and at window widths of 500 HU and 1,400 HU.

Excised Pig Lung: CTAM

Forty-three airways could be measured with both HRCT and the dissecting microscope. CT-number thresholds of 250, 300, 350, and 400 g/L were used to measure Ai. Because of the relatively greater effects of volume averaging and the poorer resolution in smaller airways, we used an airway-size-dependent threshold and SGE, based on the errors in Ai and Ao, to measure airway parameters in the pig lung specimens; a lower threshold was needed for smaller airways (i.e., using this method, the airway wall edge was more likely to be overestimated in smaller airways). Figures 7A through 7D show the absolute and relative errors for Ai and Awa. The mean errors for Ai and Awa were 0.52 ± 0.24 mm² and 0.17 ± 0.32 mm², respectively. The 95% CIs of a single measurement were therefore ± 2.22 mm² and ± 2.99 mm², respectively.

View larger version (20K): [in this window] [in a new window]   Figure 7.   Absolute (A and B) and percent (C and D) errors of measurement of Ai and Awa from excised lung, using CTAM. The x-axis is the true area measured from digitized images of the cut surface of the lung, and the y-axis is the difference between the true area and the area measured from HRCT images.

Excised Pig Lung Phantom: Manual Method

Figures 8A and 8B show the errors associated with measuring Ai and Awa, respectively, with the manual method. Ai was accurately measured, and the errors were comparable to those with CTAM; the mean error was 0.02 ± 0.15 mm² (95% CI of a single measurement: 1.11 mm²). The errors of measurement for Ai were weakly related to airway size, with a small overestimation of Ai for larger airways (r = 0.50, p = 0.01). Awa was overestimated by a mean of 6.3 ± 0.38 mm² (95% CI of a single measurement: 2.55 mm²); furthermore, the overestimation was clearly related to airway size, being greater for larger airways (r = 0.64, p < 0.01). Figures 8C and 8D show the relative errors for the measurements of Ai and Awa. Just as for to the percent errors of measurement with CTAM, the errors were largest for the smaller airways. The percent errors of measurement for Awa were clearly indirectly related to true Awa, and were also largest for the smaller airways (r = 0.89, p < 0.01). The errors with the manual method were larger than with CTAM (p < 0.01).

View larger version (18K): [in this window] [in a new window]   Figure 8.   Absolute (A and B) and percent (C and D) errors of measurement of Ai and Awa for excised lung with the manual method. The x-axis is the true area measured from digitized images of the cut surface of the lung, and the y-axis is the difference between the true area and the area measured from HRCT images. Images were photographed at a window level of 450 HU and a window width of 1,400 HU.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

In this report, we describe a novel, semiautomated CT image analysis algorithm that allows quantitative measurement of airway lumen and wall areas. The new algorithm was validated and its limits of its accuracy were defined with an airway phantom and with excised lungs that were fixed in a way that did not alter lung density. The novel aspects of this study were that: (1) the accuracy of measurement of wall area (as opposed to Ai and airway diameter in previous studies) was assessed, as well as the effects of oblique airway orientation on those measurements, (2) The optimal analysis parameters were selected through the use of inflation-fixed lungs as a gold standard. Inflation-fixed lung should be more representative of in vivo tissues than airway phantoms made of artificial materials. (3) An edge-finding algorithm (SGE) was used to determine Ao. (4) The overestimation of the airway wall edge produced by the manual method, which led to a size-dependent overestimation of Awa, was confirmed and overcome.

Previous investigators have developed algorithms that allow measurements of airway lumen and wall areas, but validation data are available only for airway-lumen measurements. Okazawa and colleagues and McNamara and associates (1, 9) described a technique in which the internal and external perimeters of the airway walls were traced on images obtained at a window width of 1,500 HU and a window level of 450 HU. They validated this technique with phantom airways made of sweet potato that were oriented perpendicularly to the angle of sectioning in the CT scan. Both Webb and colleagues and Brown and coworkers (8, 11) used algorithms that involve the operator drawing radial lines from within the airway lumen through the thickness of the airway wall.

Webb and colleagues (8) studied a bronchial phantom in which the airways were oriented at various angles. They found that the airway size, scanner collimation setting, and airway angle interact with one another, causing a variable amount of volume averaging that results in varying degrees of overestimation of the airway wall thickness, and therefore underestimation of the airway lumen diameter (Figure 9). Angulation also broadened the distribution of attenuation values across the diameter of the airway. Under such conditions, window width, as well as window level, may affect measurements. Thus, the conclusion by McNamara and colleagues (9) that window level but not window width affected measurements of luminal area and wall area was reached because these investigators only imaged tubes that were positioned parallel to the long axis of the CT scanner, thus avoiding any volume averaging. Bankier and coworkers (18) examined wall thickness along the short diameter of the airway in cadaveric lungs that were fixed in inflation and scanned using with a 3-mm collimation. They concluded that the window widths but not the window levels affected measurements of airway wall thickness; narrower window widths were associated with overestimation of this thickness. It is difficult to explain why window levels did not have any effect, and why narrow window widths were associated with overestimation rather than with underestimation in their study.

View larger version (46K): [in this window] [in a new window]   Figure 9.   Volume averaging caused by oblique orientation of an airway to the cross-sectional plane of imaging can result in an apparent decrease in the short-axis diameter of the airway lumen. The two ellipses represent the airway lumen at either end of the CT slice. Because of the angled airway orientation, the ellipses are displaced relative to each other; the distance between them corresponds to the slice thickness. The lumen appears smaller than it is because of the overlying airway walls, and even the short-axis diameter appears smaller (A) than the true short-axis diameter (B).

In the present study, a threshold method was used to define Ai and the lumen centroid was used to define the central axis of the airway so that the angle of orientation could be measured. The assumption on which this technique is based is that the central axis is represented by the centroid regardless of the airways' cross-sectional shape. This method was used by Wood and associates (14), who noted that algorithmic "leaks" during segmentation are uncommon, since there is usually high contrast between the lumen air and the airway wall. Using their three-dimensional reconstructions of the airway tree, Wood and associates were also able to make accurate measurements of cross-sectional diameter in phantom tubes larger than 2 mm in diameter and oriented at varying angles, including tubes that were oriented 90° to the long axis of the CT scanner. Their algorithm allows a three-dimensional reconstruction of the airway tree after converting rectangular CT voxels into cubes. The algorithm described in the present study is simpler, since the airway angle is determined only from three contiguous slices, and three-dimensional reconstruction is not done.

Airway-wall dimensions can be measured either as a thickness or as an area. The latter, however, is more useful when acute or chronic changes in airway wall mass are being measured, and should be used in preference to measurement of airway-wall thickness. This is because airway wall thickness increases during airway narrowing if area is conserved, so that increased wall thickness merely reflects the magnitude of narrowing. Airway wall thickness can be converted to wall area, but the calculation requires measurement of airway lumen diameter and also several assumptions. Because airways rarely appear circular, wall thickness can either be averaged from around the whole airway wall or only across the shortest diameter. The former procedure is based on the assumption that all airways are cut in cross-section, and that any noncircularity is inherent in the shape of the airway. The latter procedure is based on the assumption that airways are circular, that they remain circular during changes in caliber, and that any noncircularity is due to oblique orientation.

The results of our study confirm the importance of airway angulation in the measurement of airway luminal and wall areas. Figures 5A, 5B, 9A, and 9B show that the manual method results in an underestimation of luminal area and overestimation of wall area. The results of the study suggest that the errors are related to a constant degree of overestimation of the positions of the inner and outer airway wall edges that may be due to volume averaging of the folds along the mucosal surface and to irregularities on the adventitial surface of the airway at sites of parenchymal attachment in obliquely oriented airways. This inaccuracy in airway-wall detection leads to greater relative overestimation of Awa the smaller the airway (Figure 8D). CTAM is not associated with overestimation of Awa (Figures 4A, 4B, 8A and 8B), is unaffected by oblique airway orientation, and incorporates a correction for the angle of orientation.

CT scanners also have a built in "point-spread-function" in which the attenuation value of any pixel influences the adjacent pixels. This has the effect of enlarging small tubular structures such as blood vessels that run parallel to the CT axis, and also causes artifactual thickening of thin planar structures, such as interlobular septa and airway walls. It is this property that probably contributes to the systematic overestimation of Awa when traditional algorithms are used. This error in the measurement of Awa with the manual method, which is airway-size dependent, has implications for the reported decrease in Awa that was reported during bronchoconstriction in canine lungs (9, 11) and in normal human subjects (1). Such findings were difficult to explain; acute changes in wall area, given constant airway length, would suggest fluid shifts from the airway wall. However, increased transmural pressures during bronchoconstriction might be expected to draw fluid into the airway wall and to increase rather than decrease the wall area. Okazawa and coworkers (1) and Brown and associates (11) suspected that this apparent decrease was caused by size dependent overestimation of airway wall edge position. An overestimation of the position of the airway-wall edge would lead to a size-dependent overestimation of Awa. If Awa is conserved during airway narrowing, the smaller overestimation of wall area in the narrowed state could be interpreted as an apparent reduction in wall area. The results of the present study provide support for this explanation. CTAM, as described in this study, is more robust with respect to such measurements of Awa as a function of airway size, and it will be interesting to see it used in studies in which acute and chronic changes in Awa are anticipated.

Because it is impossible to make gold-standard measurements of Ai and Awa in vivo, we have attempted to approximate the in vivo conditions as closely as possible by studying fixed, excised lungs. The different results that we obtained with an artificial phantom and excised fixed lungs show the advantage of the latter in validating algorithms for the measurement of airway dimensions. With excised lung, the effects of volume averaging can be taken into account, whereas this cannot be done when using an airway phantom made of artificial materials, since irregularities in the adventitial and mucosal borders, varying airway orientation, thinness of the airway wall, and adjacent blood vessels cannot be duplicated in such a phantom.

In conclusion, we have developed an automated image analysis method that is reliable, fast, and easy to use for making airway measurements. Continued improvements in scanning hardware and quantitative image analysis techniques will lead to the increasing use of HRCT as a research tool. These developments may make it possible to more easily assess airway lumen and wall areas, making the development of quantitative image analysis techniques of ultimate benefit in clinical practice. Airway-wall thickening in asthmatic patients and those with chronic obstructive pulmonary disease (COPD) has been observed postmortem (19, 20) and found in descriptive HRCT studies (5, 7, 21, 22). It has also been found to contribute to airway hyperresponsiveness (AHR) in modeling studies and in morphometric studies of resected lung specimens (23). Accurate in vivo measurements of Awa could thus provide important insights into the factors responsible for AHR, asthma, and COPD. The ability to measure changes in airways over time (e.g., acutely during asthma attacks, after antiinflammatory treatment, and during growth in children) makes this technique particularly powerful.