Automated CT Scan Scores of Bronchiectasis and Air Trapping in Cystic Fibrosis

Abstract

Background:

Computer analysis of high-resolution CT (HRCT) scans may improve the assessment of structural lung injury in children with cystic fibrosis (CF). The goal of this cross-sectional pilot study was to validate automated, observer-independent image analysis software to establish objective, simple criteria for bronchiectasis and air trapping.

Methods:

HRCT scans of the chest were performed in 35 children with CF and compared with scans from 12 disease control subjects. Automated image analysis software was developed to count visible airways on inspiratory images and to measure a low attenuation density (LAD) index on expiratory images. Among the children with CF, relationships among automated measures, Brody HRCT scanning scores, lung function, and sputum markers of inflammation were assessed.

Results:

The number of total, central, and peripheral airways on inspiratory images and LAD (%) on expiratory images were significantly higher in children with CF compared with control subjects. Among subjects with CF, peripheral airway counts correlated strongly with Brody bronchiectasis scores by two raters (r = 0.86, P < .0001; r = 0.91, P < .0001), correlated negatively with lung function, and were positively associated with sputum free neutrophil elastase activity. LAD (%) correlated with Brody air trapping scores (r = 0.83, P < .0001; r = 0.69, P < .0001) but did not correlate with lung function or sputum inflammatory markers.

Conclusions:

Quantitative airway counts and LAD (%) on HRCT scans appear to be useful surrogates for bronchiectasis and air trapping in children with CF. Our automated methodology provides objective quantitative measures of bronchiectasis and air trapping that may serve as end points in CF clinical trials.

A thin-section CT scan, the best noninvasive tool for assessing bronchiectasis and structural lung injury, is commonly used to monitor cystic fibrosis (CF)-related lung disease because it provides morphologic information that is complementary to the functional information provided by lung function tests. Structural changes detected by CT scans often precede functional changes in children with CF,^1‐5 and CT scanning may be a more sensitive method than measurement of lung function for detecting disease progression.^6,7 Furthermore, CT scanning is a potential outcome measure for CF clinical trials.^8,9

To provide quantitative information, CT images must be converted into numeric data to allow statistical comparison between scans. Both qualitative visual scoring by expert readers and quantitative computer analysis have been used successfully to interpret CT scans in CF. In numerous studies, expert reader CT scan scores have been shown to correlate with other CF lung disease parameters, including clinical status and lung function.^10‐17 Although expert visual scoring such as the Brody scoring system¹⁶ includes more features of CF lung disease than can be evaluated by automated software, it is time consuming and limited by the number of expert readers available. In contrast, automated computer analysis does not require specially trained observers, avoids observer bias and variability, offers better standardization, and is a more sensitive method for detecting subtle changes.^18‐21

The most studied metric of quantitative analysis from CT scans is air trapping.²² Air trapping, a result of small airways obstruction, is indicated by areas of abnormally low lung attenuation on expiratory CT images. Air trapping has been quantified in CF and other lung diseases using a variety of lung-density approaches.^19,23‐26

Although air trapping is an important early indicator of peripheral airways disease in children with CF, the hallmark of airway pathology is bronchiectasis. The key feature of bronchiectasis on CT scanning is an enlarged airway lumen with bronchi that appear larger than the accompanying artery; however, this definition may underestimate the extent of airway damage.²⁷ Other radiographic findings include the failure of the larger airways to taper while progressing to the lung periphery and the identification of airways in the most peripheral areas of the lung, within 2 cm of the costal or paravertebral pleura.²⁸ Therefore, the number of visible airways, especially ones in the periphery, may be a surrogate of bronchiectasis on CT scans. There has been one previous report of counting visible airways in children with difficult-to-treat asthma.²⁹

The objectives of this cross-sectional pilot study were to develop and validate automated, observer-independent image analysis software to quantify airway counts and the degree of air trapping on high-resolution CT (HRCT) scans in children with CF; to compare automated CT scan measures between children with CF and age-matched disease control subjects; and to examine relationships among automated CT scan measures, Brody CT scan scores, pulmonary function measures, and sputum markers of inflammation in children with CF.

Materials and Methods

Study Subjects

During routine outpatient clinic visits, we randomly recruited 35 children between the ages of 6 and 15 years who had CF. These children were studied during times of clinical stability, defined by both clinical impression and having had no hospitalizations or changes in antibiotic regimen during the 1 month prior to being studied. Twelve disease control subjects of similar ages who had undergone an HRCT scan for a variety of clinical indications (asthma, n = 4; chronic cough, n = 2; suspected interstitial lung disease, n = 2; exercise-induced dyspnea and/or hypoxia, n = 3; and pulmonary hypertension, n = 1) were identified. Inclusion criteria for the control group were having a similar age range to that of the CF group, CT scans showing no evidence of parenchymal or airways disease, and minimal air trapping on expiratory images. The criteria used for defining the control group did not include lung function or anthropometric measurements.

The protocol, No. 03-759, was approved by the Colorado Multiple Institutional Review Board, and informed consent was obtained from each of the subjects with CF, their parents, or both. Institutional approval was obtained to analyze deidentified HRCT scans from the control subjects.

Study Design

Each subject with CF underwent an HRCT scan of the chest, pulmonary function testing, and sputum induction during one outpatient clinic visit. All HRCT scans were performed between 2003 and 2005 on an Asteion S4 scanner (Toshiba Medical Systems Europe) using a pediatric protocol with a tube potential of 120 kVp, a current of 140 mA, and a 1-s scan time. Slice thickness was 1 mm, and scans were reconstructed with a high-frequency algorithm. The image matrix was 512 × 512 pixels, and the field of view was determined based on the subject’s size. Inspiratory images were obtained at 10-mm intervals, extending from the lung apices to below the costophrenic angles, at full voluntary inspiration. During a breath hold after full voluntary exhalation, six equally spaced expiratory images were obtained. The total effective radiation dose was estimated to be < 6.23 mSv, with variation based on patient size. The HRCT scans were evaluated independently by two individuals (rater 1 and rater 2) using the Brody scoring system¹⁶ to generate overall lung disease scores and subscores (standardized, expressed as percent: 0% to 100%) for the presence and severity of bronchiectasis and air trapping.

Spirometry was performed according to American Thoracic Society guidelines^30,31 using a Sensormedics Vmax22 system (CareFusion). Lung volumes were obtained by plethysmography. The functional indexes measured included FEV₁, FVC, midvolume forced expiratory flow (FEF_25-75), total lung capacity (TLC), residual volume (RV), and RV/TLC. Percent predicted values were based on Wang et al³² and Hankinson et al.³³

Sputum was induced by having the subject inhale 3% hypertonic saline for six 2-min sessions, as described previously.³⁴ The sputum was collected into two containers, which were both transported on ice to the laboratory for processing within 20 min. One specimen was processed for quantitative cultures, according to the findings of a consensus conference on CF microbiology.³⁵ Pseudomonas aeruginosa and Staphylococcus aureus infection status was defined based on the result of this single sputum culture. The other specimen was processed for cytology and measurement of the inflammatory mediators’ free neutrophil elastase activity and matrix metalloprotease (MMP)-9, as described previously.³⁴

Image Processing and Analysis

Automated image analysis software was developed to detect and count visible airways on inspiratory CT images and to estimate the amount of air trapping on expiratory images. CT scan data were stored in DICOM format, with each pixel assigned a Hounsfield unit (HU) density value. A histogram of the HUs present on the inspiratory CT images was created and smoothed using a low-pass filter. The histogram had two large peaks, one close to −1,000 HU (indicative of air density) and one close to 0 HU (indicative of more dense structures, including soft tissue and bone) (e-Fig 1 ^{(7.4MB, pdf)} ). The software determined an air threshold value (ATV) defined as the minimum HU between these two peaks, and lung fields were segmented and isolated by selecting all pixels with an HU below the ATV. Next, the lung fields were smoothed, segmented, and divided into the left and right lungs. The total area of each lung was calculated. Discrete white objects within the lung segments, including interstitium, airway walls, mucus, and blood vessels, were retained in the lungs to create a lung mask.

The inspiratory CT scan series contained from 12 to 30 slices, depending on patient size. Slices were excluded in an automated fashion if they did not include lung fields or if the lung area they contained was too small for processing. Artifacts representing air cavities such as the stomach were removed automatically based on their location in the scan. Each inspiratory CT scan series was divided into three vertical regions defined as follows: upper slices (the top 25% of the lungs by height), center slices (slices below the top 25% and above the bottom 30% based on lung height), and lower slices (representing the remaining bottom 30% of lung slices).

The scans were further divided into peripheral and central regions. The peripheral region was defined as the outer 2 to 3 cm of lung parenchyma (“edge” of the lung), which is generally devoid of visible bronchi in healthy individuals.²⁸ From a software programming perspective, the peripheral region was first determined using the inspiratory CT scan slice closest to the carina. The peripheral region was set as the outer 10% of the maximal width of both segmented lungs through the carina (white dotted line in e-Fig 2 ^{(7.4MB, pdf)} ). Because the large airways lie close to the carina, the central regions of the center vertical slices were connected between the left and right lung (black dotted line in e-Fig 2 ^{(7.4MB, pdf)} ), creating C-shaped peripheral regions rather than ring-shaped regions. Additionally, the areas of the peripheral regions increased in proportion to the distance from the carina, because the main bronchi that emanate from the pulmonary hilum are concentrated in the center vertical slices. Therefore, the upper and lower vertical slices contained larger peripheral areas than did the center vertical slices.

Airways were selected using the following iterative algorithm. Initially, regions of interest (ROI) were selected to be areas with a density below −1,050 HU completely surrounded by higher attenuation pixels. Iterations increased the threshold by 10 HU, and new ROIs were selected and added to previous ones. Iterations were stopped once the attenuation value reached the ATV. The resulting ROIs were validated by analyzing an outline of 3 pixels surrounding each ROI. If the majority of the outline pixels had an attenuation value higher than the ATV and if they completely enclosed the ROI without gaps, then the ROI was accepted as an airway. Airways were labeled by location within the central and peripheral regions and were counted (Fig 1A). The airway count was divided by the number of inspiratory CT scan slices per scan so that the final outcome measure was the average airway count per slice.

Figure 1.

Airway counts on inspiratory slices and low attenuation density (%) on expiratory slices. A, Lung segmentation with airways in peripheral and central regions counted on inspiratory images. B, Lung segmentation with low-density areas in black on expiratory images.

On expiratory images, lung fields were isolated and segmented from the six equally spaced CT images. The software recognized air trapping by identifying areas of low attenuation next to regions with more normal attenuation. Lung attenuation histograms for the segmented lung regions were calculated. Low attenuation limits defining air trapping were set based on the attenuation in the corresponding inspiratory CT scan slices, specifically the inspiratory CT scan slice nearest to the carina. To account for variation in technique, an HU representing air trapping was calculated for each subject’s CT scan. For each patient, the lower peak of the inspiratory histogram was assumed to be the HU of fully inflated lung tissue (e-Fig 3 ^{(7.4MB, pdf)} ). The peak of the expiratory histogram shifted to higher HU density values because the lungs contained less air after exhalation. The software determined the HU density that was 10% lower than the peak of the inspiratory histogram and designated this as the threshold or cutoff value for air trapping (e-Fig 3 ^{(7.4MB, pdf)} ). The software then identified pixels on the six expiratory CT scan slices that had an HU below this cutoff value. A low attenuation density (LAD) index, defined as the total amount of air trapped in the patient’s lung as a percentage of the total lung area on the six slices, was calculated (Fig 1B). For comparison, the percentage of pixels below a fixed threshold of −850 HU was also calculated for the six expiratory scans.

Statistical Analysis

Data were analyzed using a commercially available statistical package (SAS version 9.2; SAS Institute Inc). Descriptive statistics and scatter plots were used to summarize HRCT scan scores (airway counts and LAD [%]) and lung function by study group. A log₁₀ transformation was used for data with nonnormal distributions (free neutrophil elastase activity, MMP-9 activity). Independent Student t tests were used for across-group comparisons. To compare lung zones within the CF cohort, paired comparisons were reported. Linear regression and Pearson’s correlation coefficient (r) with corresponding P values and 95% CIs were used to assess the relationship between continuous variables. A two-sided α level of 0.05 was used to determine statistical significance. Bland-Altman figures and intraclass correlation coefficients (ICCs) were generated to assess agreement across the two Brody score raters.^36,37

Results

Clinical characteristics of the children with CF (43% girls; mean age 11.1 years [range, 6-15 years]) are displayed in Table 1. The control subjects were relatively age-matched children (50% girls; mean age 10.5 years [range, 8-13 years]) who were undergoing an HRCT scan for the clinical indications listed previously.

Table 1.

—Clinical Characteristics of Subjects With CF

Characteristic	Mean (SD)	No. (%)
Female	…	15 (42.9)
Age, y	11.1 (2.6)	…
CF genotype
F508del/F508del	…	14 (40)
F508del/other	…	15 (43)
Other/other	…	6 (17)
FVC % predicted	95.5 (17.6)	…
FEV1 % predicted	95.3 (15.0)	…
FEF25-75 % predicted	84.7 (25.2)	…
RV % predicteda	128.5 (49.8)	…
RV/TLCa	27.1 (9.4)	…
Weight, percentile	37.7 (24.9)	…
Height, percentile	34.9 (25.7)	…
BMI	17.3 (2.3)	…
BMI, percentile	42.3 (25.0)	…
Pseudomonas aeruginosab	…	6 (17)
Staphylococcus aureusb	…	16 (46)

CF = cystic fibrosis; FEF_25-75 = midvolume forced expiratory flow; RV = residual volume; TLC = total lung capacity.

^aRV and RV/TLC were available from 28 of 35 subjects.

^bRespiratory culture data were available from 32 of 35 subjects (91%) who expectorated induced sputum samples for microbial analysis at their baseline visit.

Children With CF vs Control Subjects

All inspiratory HRCT scans were analyzed using the automated system. The average number of total, central, and peripheral airways counted on each inspiratory slice was significantly higher in the children with CF compared with the disease control subjects (Fig 2A, Table 2). One CF expiratory HRCT scan could not be analyzed using the automated system because of poor image quality. LAD (%) was significantly increased in the remaining subjects with CF compared with the disease control subjects (Fig 2B, Table 2). The largest difference in LAD (%) between the two groups was seen in the upper lung zones, although significant differences were present in all lung zones. In the CF cohort, LAD (%) was highest in the upper lung zone compared with the middle (mean difference, 5.6%; 95% CI, 2.3-8.8; P = .002) and lower (mean difference, 3.7%; 95% CI, 0.0-7.5; P = .052) lung zones (Table 2). There was no significant difference between the variable threshold LAD index and the fixed threshold measurement.

Figure 2.

A, Total airway counts averaged over the number of inspiratory slices. B, LAD (%) averaged over six expiratory slices. Both are increased in CF vs DC. Horizontal lines represent the group median. CF = patients with cystic fibrosis; DC = disease control subjects; LAD = low attenuation density.

Table 2.

—Automated Airway Counts and LAD (%) by Study Group

Automated CT Scan Measure	CF	Disease Control	P Value
Inspiratory airway counta
Inspiratory slices per scan, No.	16.1 (3.2)	15.9 (2.4)	.82
Total	13.2 (6.5)	7.0 (2.6)	< .0001
Central	11.4 (4.7)	6.6 (2.2)	< .0001
Peripheral	1.8 (2.0)	0.4 (0.5)	.0003
Expiratory LAD (%)b
Total	11.4 (10.6)	4.2 (2.6)	< .001
Upper lung zone	15.0 (15.6)	3.7 (4.6)	< .001
Middle lung zone	9.4 (9.9)	3.1 (2.7)	.002
Lower lung zone	11.3 (10.6)	5.5 (3.5)	.008
Expiratory fixed (%)c
Total	11.4 (10.4)	4.0 (2.4)	< .001
Upper lung zone	14.7 (15.0)	2.6 (2.6)	< .0001
Middle lung zone	9.1 (9.2)	3.3 (2.6)	.002
Lower lung zone	11.8 (11.0)	5.5 (3.3)	.004

Data are presented as mean (SD). HRCT = high-resolution CT; LAD = low attenuation density. See Table 1 legend for expansion of other abbreviation.

^aCF: n = 35; disease control: n = 12. The number of visible airways (total, central, and peripheral) averaged over all inspiratory HRCT scan slices in an individual.

^bCF: n = 34; disease control: n = 12. Total LAD (%) averaged over six expiratory HRCT scan slices in an individual subject. Upper lung zones are composed of expiratory scan levels 1 and 2; middle lung zones are expiratory scan levels 3 and 4; and lower lung zones are expiratory scan levels 5 and 6.

^cCF: n = 34; disease control: n = 12. Air trapping score calculated using fixed −850 Hounsfield unit threshold.

Automated Measures and Brody CT Scan Scores in the CF Cohort

Prior to examining the relationship between the automated CT scan measures and the Brody scores in the population with CF, the agreement between the two raters was examined. Total Brody scores ranged from 0.5% to 35.1%, air trapping subscore from 0% to 72.2%, and bronchiectasis subscore from 0% to 24.7%. Bland-Altman plots of the differences vs the average scores indicate that rater agreement was quite good for the bronchiectasis subscore (ICC = 0.88), although in four HRCT scans, the bronchiectasis subscore differed by more than five points (e-Fig 4A ^{(7.4MB, pdf)} ). The air-trapping subscores were less concordant, with the ICC = 0.58 (e-Fig 4B ^{(7.4MB, pdf)} ). For the Brody overall lung disease score, one rater tended to score higher than the other, but most of the variability in the overall score was among patients (ICC = 0.80) and was not caused by the raters (e-Fig 4C ^{(7.4MB, pdf)} ).

In the CF cohort, total and peripheral airway counts were highly correlated with bronchiectasis subscores and overall lung disease scores for both raters (Fig 3, Table 3). LAD (%) correlated significantly with Brody air trapping subscores and overall lung disease scores for both raters (Fig 4, Table 3), although the strength of the association was less than that observed between airway counts and Brody scores. LAD (%) correlated weakly with total airway counts in the CF cohort (r = 0.51; 95% CI, 0.20-0.72) (Fig 5).

Figure 3.

Peripheral airway counts correlate with Brody bronchiectasis and overall Brody scores in the CF cohort. A, Brody bronchiectasis. B, Overall Brody scores. See Figure 2 legend for expansion of abbreviation.

Table 3.

—Correlations (r) Among Automated Airway Counts, LAD (%), and Brody Scores in Subjects With CF

Automated CT Scan Measure	Rater	Air Trapping Score		Bronchiectasis Score		Overall Score
		r	95% CI	r	95% CI	r	95% CI
Total airways	1	…	…	0.82	0.67-0.91	0.79	0.62-0.89
	2	…	…	0.87	0.76-0.93	0.89	0.79-0.94
Central airways	1	…	…	0.78	0.60-0.89	0.75	0.55-0.87
	2	…	…	0.83	0.68-0.91	0.85	0.72-0.92
Peripheral airways	1	…	…	0.75	0.55-0.87	0.83	0.69-0.91
	2	…	…	0.85	0.72-0.92	0.9	0.81-0.95
LAD (%)	1	0.83	0.68-0.91	…	…	0.64	0.38-0.81
	2	0.69	0.45-0.83	…	…	0.61	0.33-0.79

See Table 1 and 2 legends for expansion of abbreviations.

Figure 4.

LAD (%) correlates with Brody air trapping and overall Brody scores in the CF cohort. A, Brody air trapping. B, Overall Brody scores. See Figure 2 legend for expansion of abbreviations.

Figure 5.

Total airway count per slice correlates with LAD (%). Solid line is estimated regression. See Figure 2 legend for expansion of abbreviation.

Airway Counts in the CF Cohort

Among the subjects with CF, visible airway counts did not differ significantly between boys and girls (mean, female = 14.0 vs male = 12.7; P = .58). However, they tended to increase with age (total airways: 0.99 airways/y, r = 0.38, P = .02; central airways: 0.75 airways/y, r = 0.40, P = .014; peripheral airways: 0.24 airways/y, r = 0.30, P = .075).

In subjects with CF, total airway counts were inversely related to FVC, with an estimated drop in FVC (% predicted) of 1.3 and 4.4 for each additional total or peripheral airway, respectively (Fig 6A, e-Table 1 ^{(7.4MB, pdf)} ). Similarly, airway counts correlated negatively with FEV₁ % predicted, with peripheral airways having the strongest relationship (Fig 6B, e-Table 1 ^{(7.4MB, pdf)} ). Airway counts did not correlate with FEF_25-75 (Fig 6C, e-Table 1 ^{(7.4MB, pdf)} ). Regarding sputum markers of inflammation, visible airway counts correlated with free neutrophil elastase activity (total airways: 0.10 log₁₀ increase in neutrophil elastase per airway, r = 0.55, P < .01; central airways: 0.12 log₁₀ increase per airway, r = 0.54, P < .01; peripheral airways: 0.31 log₁₀ increase per airway, r = 0.51, P < .01) (Fig 7A, e-Table 1 ^{(7.4MB, pdf)} ). No significant relationships were observed between airway counts and MMP-9 activity (Fig 7B, e-Table 1 ^{(7.4MB, pdf)} ). Airway counts were not significantly different between subjects who were P aeruginosa-positive and subjects who were P aeruginosa-negative (total airways: mean difference, 1.7; 95% CI, −4.6-8.1; P = .58; central airways: 1.3, 95% CI, −3.2-5.9; P = .55; peripheral airways: 0.4, 95% CI, −1.5-2.3; P = .67). Similarly, there were no significant differences in airway counts between subjects who were Staphylococcus aureus-positive and subjects who were Staphylococcus aureus-negative (total airways: mean difference, 2.2; 95% CI, −3.6-8.1; P = .44; central airways: 2.2, 95% CI, −2.0-6.4; P = .30; peripheral airways: 0.06, 95% CI, −1.7-1.9; P = .95).

Figure 6.

A, Airway counts correlate with FVC % predicted. B, Airway counts correlate with FEV₁ % predicted. C, Peripheral airway counts do not correlate with FEF 25-75. Solid line is estimated regression. FEF 25-75 = midvolume forced expiratory flow; pred = predicted.

Figure 7.

A, Airway counts correlate with Log10 of neutrophil elastase. B, Airway counts do not correlate with Log10 of MMP-9. Solid line is estimated regression. MMP-9 = matrix metalloprotease-9.

LAD (%) in the CF Cohort

In subjects with CF, LAD (%) tended to increase with age, with an estimated slope of 1.3% per year (r = 0.31, P = .07), and was marginally higher in girls (female mean = 15.7%, male mean = 8.3%, P = .08). There were no correlations found between LAD (%) (total, upper, mid-, and lower lung zones) and the lung function tests, including RV and RV/TLC, or either of the sputum markers of inflammation.

Discussion

In this cross-sectional pilot study, we demonstrated the use of fully automated image analysis software that provided quantitative measures of visible airway counts and air trapping derived from HRCT scans. Both airway counts and LAD (%) were significantly increased in the children with CF compared with disease control subjects. In the CF cohort, airway counts correlated strongly with expert reader-derived bronchiectasis subscores, and a LAD index correlated well with expert reader-derived air trapping subscores, which suggests that visible airway counts and LAD (%) may be useful surrogates of bronchiectasis and air trapping, respectively. Further, we found that total and peripheral airway counts were significantly related to lung function measurements (FVC and FEV₁) and sputum neutrophil elastase. Because neutrophil elastase has been implicated in CF lung disease pathogenesis and the development of bronchiectasis,^38‐40 the observed relationship provides further support for the use of visible airways as a surrogate for bronchiectasis.

Previous CF studies using automated CT scanning software have focused primarily on measuring lung density.^18,19,25,41 The method put forth by Goris et al²⁵ defined different severities of air trapping using information derived from both expiratory and inspiratory histograms, similar to the approach we took. In their study of patients with mild CF, all CT scan measures of air trapping correlated weakly with functional air trapping as measured by RV/TLC but did not correlate with FVC or FEV₁.²⁵ The same analysis method was used in a clinical trial of dornase alfa in children with mild CF lung disease.¹⁸ After 3 and 12 months of therapy with dornase alfa or placebo, only quantitative air trapping by CT scan showed a difference that approached statistical significance. These investigators concluded that quantitative air trapping was a more sensitive outcome measure than lung function indexes and expert-reader CT scan scores in detecting treatment effects in children with CF with mild lung disease. Quantitative airway dimensions have been examined in other CF groups and significant correlations between bronchial dilatation and lung function have been demonstrated.^42‐44 However, in contrast to our fully automated method, these studies involved semiautomated computer programs with input from expert readers to measure airway dimensions.

Building on these studies, we believe that automated image analysis can enhance the value of HRCT imaging both in CF clinical trials and in routine care. We propose that airway counts on inspiratory CT images represent large airway abnormalities, whereas an LAD index reflects small airway abnormalities. In addition, we speculate that airway counts are more likely to change over months to years, whereas an LAD index may change in a shorter time frame. A prospective study evaluating the change in these measures over time is being performed. In our study, we identified visible airway counts by their location in the central or peripheral lung zones. Peripheral airway counts correlated marginally better with Brody bronchiectasis scores and lung function than did total airway counts; however, total airway counts alone were an adequate surrogate for bronchiectasis, and this could serve to simplify the automated analysis.

The correlations between LAD (%) and Brody air trapping and overall scores were weaker than those observed between visible airway counts and Brody bronchiectasis and overall scores. LAD (%) did not correlate with either lung function or sputum inflammatory markers. Our HRCT scan protocol minimized radiation because only six expiratory slices were obtained, which may have limited the observed correlations. Two studies have shown that CT scan air trapping estimates become less precise when fewer slices are used to calculate air trapping scores.^20,25 Furthermore, an LAD index is an effort-dependent measurement, and changes in lung attenuation may result from intrasubject and intersubject variability in respiratory maneuvers. In our study, CT scans were performed without spirometric gating, which increases CT scan availability and simplifies the CT scan procedure. Controversy exists regarding the need for spirometric control of CT scans in research. In general, CT imaging studies using the Brody score have not used spirometric control. If studies are performed without spirometer control, they must be performed with attention to correct technique, including practiced and coached respiratory maneuvers by members of the radiology or pulmonology team.^45,46

Alternatively, a LAD index may measure areas that represent a different pathophysiologic process than air trapping that is measured functionally. To circumvent the problem of lung density changes with growth and to control for variable inspiration and exhalation among subjects, a computer-assisted histogram was used to identify the threshold of air trapping, using each subject’s inspiratory lung density as the internal control for expiratory thresholding. More studies are needed in subjects with a wider age range to demonstrate an advantage of this approach compared with using a fixed threshold.

Our HRCT images were thin slices obtained at intervals, rather than thin volumetric CT scans. This technique does not allow three-dimensional reconstruction and, thus, limited our analysis to only the inspiratory and expiratory slices that were available. Volumetric scans would likely facilitate the detection of specific regional changes in longitudinal studies. Volumetric measurement of air trapping has been shown to be an improvement over limited sampling methods.⁴⁷ Using a state-of-the-art CT scanner and current technique guidelines, a volumetric inspiratory and expiratory CT scan study can be performed in seconds using a dose < 2 mSv.⁴⁸ Current CT scanning technology would allow longitudinal comparison of airway counts and LAD (%).

In terms of CT scan analysis, expert scoring systems are less dependent on subtle differences in scanning techniques and are able to recognize and quantify a broader range of structural abnormalities, including mucus plugging and atelectasis. Despite these advantages, reproducibility of the Brody expert observer system has shown less than ideal agreement, particularly when subscores are analyzed.¹⁶ Expert reader scoring systems are also limited by the number of experts available and the need to retain them throughout the study period to lessen interobserver variability, which may be particularly challenging in long-term prospective CF studies.

Conclusions

In conclusion, our automated approach offers two novel quantitative CT scan measures, visible airway counts and a LAD index, that correlate well with bronchiectasis and air trapping, respectively, which are two of the most important CT scan features of structural lung disease in school-aged children with CF. These quantitative CT scan scores can complement data derived from expert scoring systems and can augment the value of HRCT scanning in CF. Quantitative CT scan analysis may prove valuable not only in CF clinical trials, but also in routine clinical care, to monitor changes in structural lung disease.

Supplementary Material

Online Supplement

Abbreviations

ATV

air threshold value

CF

cystic fibrosis

FEF_25-75

midvolume forced expiratory flow

HRCT

high-resolution CT

HU

Hounsfield unit

ICC

intraclass correlation coefficient

LAD

low attenuation density

MMP

matrix metalloprotease

ROI

region of interest

RV

residual volume

TLC

total lung capacity