A comparison of visual and quantitative methods to identify interstitial lung abnormalities

Abstract

Background: Evidence suggests that individuals with interstitial lung abnormalities (ILA) on a chest computed tomogram (CT) may have an increased risk to develop a clinically significant interstitial lung disease (ILD). Although methods used to identify individuals with ILA on chest CT have included both automated quantitative and qualitative visual inspection methods, there has been not direct comparison between these two methods. To investigate this relationship, we created lung density metrics and compared these to visual assessments of ILA.

Methods: To provide a comparison between ILA detection methods based on visual assessment we generated measures of high attenuation areas (HAAs, defined by attenuation values between -600 and -250 Hounsfield Units) in >4500 participants from both the COPDGene and Framingham Heart studies (FHS). Linear and logistic regressions were used for analyses.

Results: Increased measures of HAAs (in ≥ 10 % of the lung) were significantly associated with ILA defined by visual inspection in both cohorts (P < 0.0001); however, the positive predictive values were not very high (19 % in COPDGene and 13 % in the FHS). In COPDGene, the association between HAAs and ILA defined by visual assessment were modified by the percentage of emphysema and body mass index. Although increased HAAs were associated with reductions in total lung capacity in both cohorts, there was no evidence for an association between measurement of HAAs and MUC5B promoter genotype in the FHS.

Conclusion: Our findings demonstrate that increased measures of lung density may be helpful in determining the severity of lung volume reduction, but alone, are not strongly predictive of ILA defined by visual assessment. Moreover, HAAs were not associated with MUC5B promoter genotype.

Fig. 1

A density plot of the percentage of the lung occupied by High Attenuation Areas (HAAs, chest CT attenuation values between −600 and −250 HU) in participants with ILA (in red, n = 163), in participants indeterminate for ILA (in gray, n = 757), and in the participants without ILA (in black, n = 1173). Despite the differences in numbers between the groups for each category (defined by color) the area under the curve is normalized to a density of 1 which gives a sense of the relative spread of the data between categories. The percentage of lung occupied by various HAA thresholds is listed on the x-axis. The density of subjects at various HAA thresholds is listed on the y-axis

Fig. 2

On the vertical axis we present representative examples of (a) a subject with interstitial lung abnormalities identified by visual assessment but has less than 10 % of the lung with chest CT attenuation values between −600 and −250 HU (HAA) (7.5 % HAA) and (b) a subject having > 10 % of the lung with chest CT attenuation values between −600 and −250 HU (HAA) (10.8 %) but not identified as having interstitial lung abnormalities identified by visual assessment. Each row represents data from a single subject. On the horizontal axis we present axial high resolution chest computed tomographic (HRCT) images (1 [approximately at the level of the carina] and 2 [approximately at the level of the right inferior pulmonary vein])