Purpose: To develop a system for texture-based quantification of emphysema on high-resolution computed tomography (HRCT) and to compare it with density-based quantification in correlation with pulmonary function test (PFT).
Materials and methods: Two hundred sixty-one circular regions of interest (ROI) with 16-pixel diameter [66 ROIs representing typical area of normal lung; 69 representing bronchiolitis obliterans (BO); 64, mild emphysema (ME); and 62, severe emphysema (SE)] were used to train the automated classification system based on the Support Vector Machine classifier and on variable texture and shape features. An automated quantification system was developed with a moving ROI in the lung area, which classified each pixel into 4 categories. To validate the system, the HRCT and standard-kernel-reconstructed volumetric CT data of 39 consecutive patients with emphysema were included. Using this system, the whole lung area was evaluated, and the area fractions of each class were calculated (normal lung%, BO%, ME%, SE%, respectively). The emphysema index (EI) of texture-based quantification was defined as follows: (0.3 x ME% + SE%) (TEI). EIs from density-based quantification with a threshold of -950 Hounsfield Units, were measured on both HRCT (DEI_HR_2D) and on volumetric CT (DEI_standard_3D). The agreement between TEI, DEI_HR_2D, and DEI_standard_3D was assessed using interclass correlation coefficients (ICC). Correlation of the results on the TEI with the PFT results was compared with the results of the DEI_standard_3D and the DEI_HR_2D with Spearman's correlation test. To evaluate the contribution of each texture-based quantification lesion (BO%, ME%, SE%) on PFT, multiple linear regression analysis was performed.
Results: The calculated TEI (19.71% +/- 17.98%) was well correlated with the DEI_standard_3D (19.42% +/- 14.30%) (ICC = 0.95), whereas the ICC with DEI_HR_2D (37.22% +/- 9.42%) was 0.43. TEI showed better correlation with PFT than DEI_standard_3D or DEI_HR_2D did [R = 0.71 vs. 0.67 vs. 0.61 for forced expiratory volume in 1 second (FEV(1))/forced vital capacity (FVC); 0.54 vs. 0.50 vs. 0.43 for diffusing capacity (DLco), respectively]. Multiple linear regression analysis revealed that the BO% and SE% areas were independent determinants of FEV(1)/FVC, whereas the ME% and the SE% were determinants of DLco.
Conclusion: Texture-based quantification of emphysema using an automated system showed better correlation with the PFT results than density-based quantification. Separate quantification of the BO, ME, and SE areas showed a different contribution of each component to the FEV(1)/FVC and the DLco. The proposed system can be successfully used for detailed regional and global evaluation of lung lesions on HRCT scanning for emphysema.