CT-derived Biomechanical Metrics Improve Agreement Between Spirometry and Emphysema

Abstract

Rationale and Objectives

Many COPD patients have marked discordance between FEV₁ and degree of emphysema on CT. Biomechanical differences between these patients have not been studied. We aimed to identify reasons for the discordance between CT and spirometry in some patients with COPD.

Materials and Methods

Subjects with GOLD stage I–IV from a large multicenter study (COPDGene) were arranged by percentiles of %predicted FEV₁ and emphysema on CT. Three categories were created using differences in percentiles: Cat_spir with predominant airflow obstruction/minimal emphysema, Cat_CT with predominant emphysema/minimal airflow obstruction, and Cat_matched with matched FEV₁ and emphysema. Image registration was used to derive Jacobian determinants, a measure of lung elasticity, anisotropy and strain tensors, to assess biomechanical differences between groups. Regression models were created with the above categories as outcome variable, adjusting for demographics, scanner type, quantitative CT-derived emphysema, gas trapping, and airway thickness (Model 1), and after adding biomechanical CT metrics (Model 2).

Results

Jacobian determinants, anisotropy and strain tensors were strongly associated with FEV₁. With Cat_matched as control, Model 2 predicted Cat_spir and Cat_CT better than Model 1 (Akaike Information Criterion, AIC 255.8 vs. 320.8). In addition to demographics, the strongest independent predictors of FEV₁ were Jacobian mean (β= 1.60,95%CI = 1.16 to 1.98; p<0.001), coefficient of variation (CV) of Jacobian (β= 1.45,95%CI = 0.86 to 2.03; p<0.001) and CV strain (β= 1.82,95%CI = 0.68 to 2.95; p = 0.001). CVs of Jacobian and strain are both potential markers of biomechanical lung heterogeneity.

Conclusions

CT-derived measures of lung mechanics improve the link between quantitative CT and spirometry, offering the potential for new insights into the linkage between regional parenchymal destruction and global decrement in lung function in COPD patients.

Introduction

The diagnosis of chronic obstructive pulmonary disease (COPD) is currently based on the detection of airflow obstruction by spirometry.¹ It is increasingly recognized that airflow obstruction as measured by impairment in the forced expiratory volume in 1 second (FEV₁) does not fully explain the morbidity associated with the disease, and this functional definition can be complemented by anatomic measures of disease using widely available imaging modalities.² Computed tomography (CT) has become the gold standard in the quantitative assessment of the presence and distribution of emphysema, a major component of COPD, and relies on using a fixed Hounsfield threshold value below which all lung areas are deemed emphysematous in a CT scan obtained at full inspiration.³ CT measures of emphysema correlate well with pathology,⁴ and numerous studies have shown a strong correlation between spirometry and CT emphysema. ^5–12 The agreement between CT emphysema and spirometry is however not perfect, and in some cases, CT densitometry may be more sensitive in detecting emphysema than spirometry.^6,13

It is our observation that many COPD patients have marked discordance between FEV₁ and degree of emphysema on volumetric CT.^14,15 Some subjects with severe airflow obstruction have mild emphysema on CT and conversely, some patients with severe emphysematous destruction of the lung have relatively mild spirometric impairment. While some of these differences, especially in the former group, are likely due to airway narrowing, the reasons for this discrepancy between expected changes on spirometry and CT have not been systematically studied, particularly in the disproportionate emphysema group. Since airflow obstruction is due to a combination of airway narrowing and loss of elastic recoil due to emphysema, it is possible that static single-volume CT images do not capture lung mechanics sufficiently to explain lung function defects. We hypothesized that biomechanical measures of regional lung tissue expansion and contraction using image registration applied to paired inspiratory and expiratory CT scans will provide a link between CT-derived quantitative measures and spirometry. Through a demonstration of this link, we seek to provide an improved understanding of patient specific links between the presence and distribution of quantitative emphysema and airflow obstruction.

Materials and Methods

Data Collection

Data for this study was acquired from the Genetic Epidemiology of COPD (COPDGene) study; this is a large multicenter study of current and former smokers aged 45 to 80 years. Details of the study protocol have been previously published.¹⁶ Post bronchodilator spirometry was performed using the ndd Easy-One spirometer to assess airflow obstruction.¹⁷ COPD was diagnosed based on a fixed threshold for the ratio of FEV₁ to the forced vital capacity (FVC) of <0.70; disease severity was graded according to the Global initiative for chronic Obstructive Lung Disease (GOLD) guidelines.¹ Reference values for spirometry were drawn from the National Health and Nutrition Examination Survey (NHANES) III cohort.¹⁸ Volumetric CT scans were acquired with the subject in supine position during a carefully coached breath hold to either full inspiration (total lung capacity, TLC) or end tidal expiration (functional residual capacity, FRC); or at one center to full expiration (residual volume, RV).¹⁶ The scans followed an imaging protocol with collimation, 0–5mm; tube voltage, 120kV; tube current 200mAs; gantry rotation time of 0.5s; and pitch, 1.1. The images were reconstructed with a standard kernel with a slice thickness of 0.75 mm and a reconstruction interval of 0.5 mm. 3D Slicer software (www.airwayinspector.org) was used to measure emphysema and gas trapping. Pulmonary Workstation 2 (VIDA Diagnostics, Coralville, IA, USA) was used to measure airway dimensions. ¹⁶. Emphysema was quantified by using the percentage of voxels at TLC with attenuation less than −950 Hounsfield Units (HU) (low attenuation area, %LAA950_insp), and also as the HU value at the 15th percentile (Perc15). ^16,19 Gas trapping was calculated as the percentage of voxels at FRC with attenuation less than −856 HU (%LAA856_exp).²⁰ We used wall area percentage of segmental airways (WA%) and gas trapping to measure airway disease.¹⁶ The COPDGene study was approved by the institutional review boards of all 21 participating centers, and written informed consent was obtained from each subject.

Case Selection

Subjects with GOLD stage I to IV disease, without physician-diagnosed bronchial asthma and with good quality expiratory images, were included. As there is no published data on the degree of emphysema to be expected for a given value of FEV₁, we used the percentile method to assess discrepancy between CT and spirometry to derive a sample size of convenience and also to create two widely disparate groups by spirometry and CT emphysema. In addition, as we have previously demonstrated strong independent correlations between CT emphysema, gas trapping and Jacobian metrics with airflow obstruction on spirometry, we did not analyze all subjects in the cohort as adding biomechanical metrics to measurements of lung disease with such strong correlations is not expected to provide additional explanations of mechanics, especially in subjects who are highly concordant for CT and spirometry and constitute the majority of subjects.^14,21 In order to study those who have significant discordance in the severity of CT and spirometric findings, we used the percentile method to express differences and discordance. We arranged all subjects (n=2982) in ascending order of severity of airflow obstruction as assessed by percent predicted FEV₁ to calculate the percentile values for spirometric abnormality; a second list was created by arranging all subjects in ascending order of severity of CT emphysema as determined by the continuous measure Perc15 percentile. Concordance between spirometry and CT was assessed by subtracting percentile ranking for CT from percentile ranking for spirometry. Those with values closest to 0 (n=100) were considered “Matched” as they had expected ranking for spirometry compared to CT emphysema, and this group served as the reference group. Those with the greatest positive percentile ranking difference for spirometry compared to CT (n=100) were considered spirometry predominant, and those with the greatest positive percentile ranking difference for CT compared to spirometry (n=100) were considered CT predominant. 3 subjects in the CT predominant group were excluded due to failure of image registration matching.

Image Registration

The inspiratory and expiratory CT images were registered for each subject, and expiratory image matched to inspiratory image. Details of image registration are provided in the Supplement. Briefly, a non-rigid lung mass-preserving registration method was used to capture volume changes between these two targeted lung inflation levels.^21,22 A sum of squared tissue volume difference was used as a similarity metric. This similarity criterion aims to find a registration transformation that minimizes the local difference of tissue volume inside the lungs scanned at different pressure levels. This method has been shown to be effective in lung image registration protocols.^21–23 The final transformation matrix from inspiration to expiration was derived from the registration protocol and in turn used to extract displacement field information.

Three measures were calculated from the registration process: Jacobian, strain information, and anisotropic deformation index (ADI). Jacobian measures the local volume change and estimates the pointwise volume expansion and contraction during the deformation from inspiration to expiration. Jacobian has values from 0 to infinity. A Jacobian value of 1 indicates neither local expansion nor contraction. Values greater than or lesser than 1 represent local expansion and contraction respectively. Maximum principle strain was computed from the displacement field information to extract how much a given displacement differs locally from inspiration to expiration. Strain analysis expresses the geometric deformation caused by action of stress in the lung. ADI provides the orientation preference of lung deformation by calculating the ratio of length in the direction of maximal extension to the length in the direction of minimal extension within a unit volume of 1 mm.²³ The larger the ADI value, more the anisotropy is with deformation. The coefficient of variation (CV) across the whole lung was calculated for each of these biomechanical measures to assess heterogeneity and dispersion.

Statistical Analyses

Univariate regression analyses were performed for CT variables with FEV₁ to assess association, and those with p<0.05 were included in multivariable models. Multivariable analyses were also adjusted for clinically important variables including age, sex, race, body mass index (BMI) and scanner type. Multicollinearity diagnostics were performed and variables with a variance inflation factor of greater than 10 (strain mean, ADI mean and ADI CV) were excluded from the model. Independent contribution of each covariate to the variance in FEV₁ was calculated using squared semi-partial correlation coefficient (r²).

Further, to assess differences in the discordant groups of interest, three categories were created using differences in percentile rankings: Cat_spir with predominant airflow obstruction on spirometry and minimal CT emphysema; Cat_CT with predominant CT emphysema and relatively minimal airflow obstruction on spirometry; and Cat_matched with matched FEV₁ and CT emphysema. Two regression models were created with mentioned categories as outcome variables using multinomial logistic regression. CT variables significantly associated with FEV₁ were entered into the models and all models were adjusted for age, race, sex, BMI, and CT scanner type. At this stage, %LAA950_insp was substituted for Perc15 as a measure of emphysema as Perc15 was already used to derive the groups, and %LAA950_insp is more commonly used clinically to quantify CT emphysema. Model A comprised of %LAA950_insp, %LAA856_exp and WA%. Biomechanical measures (mean Jacobian, CV of Jacobian, and CV of strain) were added to the predictors in Model A to create Model B. With Cat_matched as reference category, multinomial logistic regression was used to predict Cat_spir and Cat_CT. Model fit statistics are shown in terms of Akaike Information Criterion (AIC) derived from information theory. AIC is used to estimate the quality and model comparisons and is defined as: AIC = −2Lm+ 2k, where Lm represents maximum log-likelihood and K is the number of variables in the model. AIC takes both goodness of fit and number of variables into account while penalizing the increase in number of variables and thus avoids over fitting scenarios. The smaller the AIC, the better is the model prediction. All analyses were performed using R statistical software (Version 3.0.1) and Statistical Package for the Social Sciences (SPSS 22.0, SPSS Inc., Chicago, IL, USA).

Results

The three categories were well separated by percentile differences for spirometry and CT emphysema. The percentile difference in the Cat_spir category was −65.7 (SD8.2), range −93.5 to −53.2; difference in Cat_CT was 61.1 (7.8), range 51.6 to 79.1; and in Cat_matched was −0.04 (0.63), range −1.03 to 1.06 (p<0.001 for all comparisons). Representative cases are depicted in Figure 1. Baseline demographics, spirometry and CT features for the three categories are described in Table 1. Compared to those in the Cat_matched, those with Cat_spir were younger and were more obese. There were significant differences in airway disease between the three categories, with the disproportionate spirometric category showing most airway disease.

Figure 1.

Panel A shows computed tomographic (CT) images for subject with severe airflow obstruction (FEV₁ %predicted 32.6) but with relatively minimal emphysema (1.5% volume <−950HU on end inspiratory images). Panel B shows features for subject with severe emphysema on CT (20.8%) but with relatively minimal airflow obstruction (FEV₁ %predicted 99.6). Top row represents the overlay of emphysema voxels on the CT images. Middle row represents the overlay of Jacobian color map on the CT images from each category. Jacobian value (=1) represents no deformation; >1 represents local expansion; <1 local contraction. Bottom row represents 3D visualization of emphysema voxels in each category with flow volume loop.

Table 1.

Demographic information, radiographic and spirometry measures

	Catspir (n=100)	CatCT (n=97)	Catmatched (n=100)
Age (years)	60.4 (7.8)**	65.0 (8.2)	64.5 (9.0)
Sex (%Males)	60 (60)	69 (71)	57 (57)
Race (%Non Hispanic Whites)	74 (74)*	85 (88)	84 (84)
BMI (kg/m2)	33 (7.3)***	26.6 (5.2)	26.4 (6.0)
Smoking packyears	57.3 (29.0)	51.4 (25.8)	50.4 (24.3)
FEV1 (L)	1.15 (0.35)**	2.75 (0.68)¥	1.45 (0.82)
FEV1 % predicted	38.0 (9.0)¥	92.6 (13.7)¥	50.8 (26.6)
FVC (L)	2.29 (0.63)¥	4.48 (0.94)¥	3.01 (0.94)
FEV1/FVC	0.51 (0.10)*	0.61 (0.07)¥	0.46 (0.17)
%Emphysema (LAAinsp<−950 HU)	1.5 (1.2)¥	17.3 (8.9)	19.4 (18.0)
%Gas trapping (LAAexp<−856 HU)	20.5 (12.6)¥	34.9 (11.8)**	44.5 (26.4)
Wall Area%	65.1 (2.7)¥	59.3 (2.4)¥	62.2 (2.8)

All values expressed as mean (standard deviation) unless other specified.

^*p<0.05

^**p<0.01

^¥p<0.001

Cat_spir = category with disproportionate spirometric abnormality. Cat_CT = category with disproportionate CT abnormality. Cat_matched = matched CT and spirometric abnormalities. BMI = Body mass index. FEV₁ = Forced expiratory volume in the first second. FVC = Forced vital capacity. LAA_insp<−950 HU = Low attenuation areas <−950 Hounsfield Units at end inspiration. LAA_exp<−856 HU = Low attenuation areas <−856 Hounsfield Units at end expiration. Wall Area% = (wall area/total bronchial area)×100, calculated as the average of six segmental bronchi in each subject.

A number of CT metrics of lung deformation were associated with airflow obstruction. On univariate analyses, there was a significant association between FEV₁ and Jacobian mean (regression coefficient β = 2.49, 95%CI 2.17 to 2.81; p<0.001), Jacobian CV (β = 3.90, 95%CI 3.38 to 4.42; p<0.001), ADI mean (β = 0.25, 95%CI 0.20 to 0.30; p<0.001), ADI CV (β = 0.69, 95%CI 0.52 to 0.87; p<0.001), strain mean (β = 3.55, 95%CI 3.16 to 3.94; p<0.001) and strain CV (β = −1.68, 95%CI −3.06 to −0.30; p = 0.017). Of these, after assessment of multicollinearity, Jacobian mean, Jacobian CV and strain CV were selected for inclusion in the multivariable model to predict independent associations with FEV₁ (Table 2). We also assessed the relative independent contributions of the CT metrics and found that the Jacobian mean explained 5.6% of the variation in FEV₁ compared to CT emphysema which explained only 2.8%. Wall area% and Jacobian CV were other strong predictors, explaining 3.9% and 2.3% of the variance in FEV₁, respectively (Supplemental Table 1).

Table 2.

Univariate and Multivariable linear regression for prediction of FEV₁

	Univariate			Multivariable
Variable	Parameter estimate	95% CI	P value	Parameter estimate	95% CI	p value
Age (years)	−0.010	−0.022 to 0.003	0.134	−0.19	−0.026 to −0.011	<0.001
Male Sex	−0.69	−0.90 to −0.48	<0.001	−0.53	−0.65 to −0.40	<0.001
LAA insp<−950 HU	−0.010	−0.018 to −0.003	0.007	−0.026	−0.035 to −0.016	<0.001
LAAexp<−856 HU	−0.014	−0.019 to −0.009	<0.001	0.006	0.001 to 0.014	0.115
Wall Area%	−0.21	−0.26 to −0.16	<0.001	−0.07	−0.10 to −0.05	<0.001
Jacobian Mean	2.49	2.17 to 2.81	<0.001	1.60	1.16 to 1.98	<0.001
Jacobian CV	3.90	3.38 to 4.42	<0.001	1.45	0.86 to 2.03	<0.001
Strain CV	−1.68	−3.06 to −0.30	0.017	1.82	0.68 to 2.95	0.001

FEV₁ = Forced expiratory volume in the first second. CI = Confidence intervals. LAA_insp<−950 HU = Low attenuation areas <−950 Hounsfield Units at end inspiration. LAA_exp<−856 HU = Low attenuation areas <−856 Hounsfield Units at end expiration. Wall Area% = (wall area/total bronchial area)×100, calculated as the average of six segmental bronchi in each subject. CV = Coefficient of variation.

Multivariable model included variables significant on univariate analyses in the Table and were also adjusted for age, body mass index, sex, race and scanner type. R² for multivariable model = 0.73

Table 3 shows a comparison of measures of lung mechanics between the three categories. The same measures of lung mechanics that were significant on univariate regression with FEV₁, and after collinearity adjustments were included in the two multinomial logistic regression models shown in Table 4. Comparison of AIC between the two models shows that inclusion of biomechanical measures (Model 2) predicts the categories better than the model that contains only static single-volume based CT metrics of structural lung disease (Model 1), AIC 255.8 vs. 320.8. As worsening disease can result in increase in TLC, which in turn can compensate for increase in RV and tend to preserve FVC and thus FEV₁, we also performed sensitivity analyses with addition of CT measured TLC to the above model, with no change in prediction of FEV₁ (data not shown).

Table 3.

Biomechanical CT measures for the three categories

	Catspir (n=100)	CatCT (n=97)	Catmatched (n=100)
Jacobian Mean	1.38 (0.18)	1.73 (0.20)¥	1.44 (0.23)
Jacobian CV	0.21 (0.08)*	0.46 (0.16)¥	0.25 (0.10)
Strain Mean	0.36 (0.12)*	0.65 (0.15)¥	0.41 (0.16)
Strain CV	0.57 (0.07)¥	0.61 (0.06)	0.62 (0.09)
ADI Mean	1.03 (0.53)	3.20 (2.71)¥	1.34 (0.94)
ADI CV	1.06 (0.36)*	1.71 (0.69)¥	1.25 (0.36)

All values expressed as mean (standard deviation) unless other specified.

^*p<0.05

^¥p<0.001

CT = computed tomography. Cat_spir = category with disproportionate spirometric abnormality. Cat_CT = category with disproportionate CT abnormality. Cat_matched = matched CT and spirometric abnormalities. CV = Coefficient of variation. ADI = Anistropic deformation index.

Table 4.

Multivariable Logistic Regression Models for Predicting Disproportionate Categories

Model 1
	Catspir		CatCT
	Odds Ratio	95%CI	Odds Ratio	95%CI
LAA<−950insp	0.40¥	0.25–0.62	0.94	0.91–1.03
LAA<−856exp	1.14**	1.06–1.23	1.05*	1.01–1.10
WA%	1.96¥	1.53–2.52	1.47¥	1.24–1.75
Model 2
	Catspir		CatCT
	Odds Ratio	95%CI	Odds Ratio	95%CI
LAA<−950insp	0.62*	0.39–0.98	0.96	0.89–1.04
LAA<−856exp	0.94	0.84–1.06	0.98	0.92–1.04
WA%	1.63¥	1.22–2.18	1.20	0.98–1.47
Jacobian mean	17.05¥	4.84–60.11	4.38¥	2.15–8.93
Jacobian CV	21.98*	4.47–65.04	5.96¥	2.75–12.9
Strain CV	2.16	0.76–6.15	1.83*	1.01–3.31

Cat_spir = category with disproportionate spirometric abnormality. Cat_CT = category with disproportionate CT abnormality. Cat_matched = matched CT and spirometric abnormalities. LAA_insp<−950 HU = Low attenuation areas <−950 Hounsfield Units at end inspiration. LAA_exp<−856 HU = Low attenuation areas <−856 Hounsfield Units at end expiration. Wall Area% = (wall area/total bronchial area)×100, calculated as the average of six segmental bronchi in each subject. CV = Coefficient of variation.

All models adjusted for age, race, sex, body mass index and scanner type.

^*p<0.05

^**p<0.01

^¥p<0.001

R² for Model 1 =0. 56. R² for Model 2 = 0. 66. AIC for Model 1 = 320.8, AIC for Model 2 = 255.8.

Discussion

We show that using dual-volume based biomechanical measures of lung tissue deformation rather than static single-volume measures of CT emphysema considerably improves prediction of spirometric airflow obstruction and also concordance between CT and spirometry. This improved linkage between CT metrics and spirometry provides a validation of the mechanics-based measures derived from image matching, and offers the ability to link regional lung mechanics to spirometry which provides only a single metric reflecting a composite of regional lung differences. By providing a comprehensive map of regional lung mechanics coupled with regional maps of emphysema, patient selection for therapies for severe emphysema can be better determined and outcomes can be evaluated in light of this new understanding of lung structure-function relationships.

A number of studies have analyzed the correlation between static single-volume CT measures of emphysema and spirometric airflow obstruction. One large study examining this showed a correlation between the two of −0.76 when emphysema was defined at <−950HU threshold.²⁴ While this degree of correlation is good, it leaves room for a significant amount of discrepancy. Emphysema is a heterogeneous disease and the distribution of lung disease affects spirometry. Studies that examined the relative distribution of emphysema by zone found that upper zone emphysema correlates better with the diffusing capacity of carbon monoxide whereas predominant lower zone emphysema correlates better with FEV₁. ^25–28 We previously showed that emphysema like changes in the right middle lobe correlate the least with spirometry.²⁹ There is also a differential effect of central versus peripheral involvement of the lung, with a greater correlation of central involvement with FEV₁.²⁵ However, these studies do not provide a composite measure of emphysema that improves agreement with spirometry. While CT gas trapping as an indirect measure of small airway disease has a greater correlation with airflow obstruction,¹⁴ this is likely significantly influenced by the degree of baseline emphysema in the preceding respiratory cycles.³⁰

We add to the literature by showing three additional measures that improve agreement between CT and spirometry. Not surprisingly, the spirometry predominant group had significantly greater airway disease. Perhaps of more interest is the CT predominant group in which subjects had substantial degrees of emphysema with relatively minimal airflow obstruction. The divergence in emphysema measures and expected spirometry impairment is likely partly due to the fact that single-volume-based CT measures are static whereas spirometry is a dynamic measure. By using image registration applied to matched pairs of inspiratory-expiratory CT scans, we show an improved agreement between dynamic spirometric measures and biomechanical CT measures. While measures of airways disease including differences in airway wall thickness and air trapping also account for some of the disagreement, we found that after adjustment for indices of airway disease, biomechanical measures account for considerable additional variability. Our findings suggest that the presence of CT emphysema may not always translate into airflow obstruction, and that not all emphysema translates into loss of elastic recoil equally. The image-matching-based metrics more directly link CT-based findings with the integrated mechanics associated with spirometry.

We found that a greater value for the Jacobian mean predicts a higher FEV₁. The Jacobian mean is reflective of the over-all volume change of the individual lung regions. Ju et al showed that greater degree of lobar heterogeneity of emphysema on single-volume CT images was associated with less airflow obstruction.³¹ We extend these findings by demonstrating that image-matching based measures of biomechanical heterogeneity (Jacobian and strain CVs) are predictive of lung function, and these biomechanical changes are provided on a lobar and sub-lobar basis. It is pertinent to note that the Jacobian CV and strain CV are more strongly associated with the prediction of the Cat_CT category, one in which there is significant CT emphysema but the spirometric abnormality remains relatively masked. Though we did not assess the type of emphysema qualitatively, panlobular emphysema tends to be more homogeneous, is associated with higher lung compliance, and is associated with a lower FEV₁ for the same level of quantitative emphysema of centrilobular distribution.^32,33 Centrilobular emphysema tends to be more heterogeneous and is associated with lesser FEV₁ reduction for a given degree of emphysema. This is especially relevant for subjects with very mild airflow obstruction who might harbor significant structural lung disease prior to development of spirometric abnormality. We speculate that more local heterogeneity indicates relative canceling effects of the pressure effects created by greater and lesser expansive lung regions, a form of local pseudorestriction. The Jacobian CV might be a novel measure of biomechanical lung heterogeneity that can explain a significant proportion of the discrepancy. We acknowledge that some of the biomechanical measures that have significant associations with outcome categories have wide confidence intervals, perhaps reflecting the relatively small sample size of subjects included in the analysis.

Our findings have a number of potential clinical implications. The diagnosis of COPD has traditionally relied on demonstrating airflow obstruction on spirometry. CT emphysema has been proposed as a new metric of disease that provides complementary information but it has been observed that the amount of CT-based emphysema does not directly translate to the degree of airflow obstruction.³⁴ By providing CT-derived measures of regional lung mechanics to the more commonly used quantitative CT metrics derived from a single lung volume, we now provide a closer link between quantitative CT and spirometry. Further research is needed to find out whether these metrics can aid earlier diagnosis of disease before the onset of airflow obstruction which is a global measure of lung disease. CT emphysema is used for assessment of structural lung disease and assessing regional heterogeneity of emphysema for interventions such as bronchoscopic lung volume reduction. Patient selection might be better served by the inclusion of regional measures of lung mechanics to the list of measures assessed by quantitative CT. Addition of biomechanical measures prior to lung resection surgeries may also improve prediction of postoperative lung function.

Our study has some limitations. First, CT scans were not spirometrically gated and variations in respiratory effort may affect the reproducibility of Jacobian measures; however, patients were coached to maximum inspiration and end exhalation. Second, this was a multicenter study and hence a number of scanners were used to acquire the CT scans. However, we did adjust for scanner variability in our analyses, and emphysema and air trapping were assessed from the same lung volumes as those used for the biomechanical measures, thus linking these measures to each other, accounting, in part, for protocol differences and subject variability. Third, although we adjusted for airway disease by use of WA%, the current resolution of CT limits visualization of airways beyond the segmental level. There is also a growing understanding that WA% is a composite of changes in airway wall thickness and luminal dimensions and may not fully reflect peripheral airway disease.³⁵ To address these issues, we also used gas trapping as a surrogate of small airways disease. Fourth, CT scans were obtained at only two volumes, thus limiting our ability to account for regional differences in lung mechanics reflected in the non-linearity of the pressure volume curve either on a global or regional basis.³⁶ With the considerable reduction in radiation doses afforded by evolving CT detector and x-ray gun technologies coupled with improved iterative reconstruction methods,^37,38 improved details of mechanical characteristics of the lung, by utilizing dynamic imaging or greater numbers of lung volumes, may become more practical while limiting radiation exposure. Finally, even though we reduced the sample size to 300, this was intentionally done to examine the cases with the most discordance. This was necessary as there is no previous literature to guide us as to how much abnormality on CT predicts the abnormality on spirometry and vice versa. Although subjects with no airflow obstruction can have a significant degree of emphysema, we did not include these subjects as they are more likely to have matched CT-Spirometry values than those with disease, and hence would heavily influence our analyses. Our study also has a number of strengths. Sites were continuously coached in regards to the proper performance of lung volume coaching and CT protocol adherence. Study subjects were drawn from a cohort that is well characterized phenotypically, and hence included a large sample size and included a high proportion of African Americans.

In conclusion, compared to single-volume CT assessment of emphysema, biomechanical measures derived from dual-volume CT show improved agreement with airflow obstruction on spirometry. This has implications for disease detection, for the understanding of links between regional lung disease and spirometrically derived lung function, as well as therapy planning.