Latent traits of lung tissue patterns in former smokers derived by dual channel deep learning in computed tomography images

Frank Li, Jiwoong Choi, Chunrui Zou, John D Newell Jr, Alejandro P Comellas, Chang Hyun Lee, Hongseok Ko, R Graham Barr, Eugene R Bleecker, Christopher B Cooper, Fereidoun Abtin, Igor Barjaktarevic, David Couper, MeiLan Han, Nadia N Hansel, Richard E Kanner, Robert Paine 3rd, Ella A Kazerooni, Fernando J Martinez, Wanda O'Neal, Stephen I Rennard, Benjamin M Smith, Prescott G Woodruff, Eric A Hoffman, Ching-Long Lin, Frank Li, Jiwoong Choi, Chunrui Zou, John D Newell Jr, Alejandro P Comellas, Chang Hyun Lee, Hongseok Ko, R Graham Barr, Eugene R Bleecker, Christopher B Cooper, Fereidoun Abtin, Igor Barjaktarevic, David Couper, MeiLan Han, Nadia N Hansel, Richard E Kanner, Robert Paine 3rd, Ella A Kazerooni, Fernando J Martinez, Wanda O'Neal, Stephen I Rennard, Benjamin M Smith, Prescott G Woodruff, Eric A Hoffman, Ching-Long Lin

Abstract

Chronic obstructive pulmonary disease (COPD) is a heterogeneous disease and the traditional variables extracted from computed tomography (CT) images may not be sufficient to describe all the topological features of lung tissues in COPD patients. We employed an unsupervised three-dimensional (3D) convolutional autoencoder (CAE)-feature constructor (FC) deep learning network to learn from CT data and derive tissue pattern-clusters jointly. We then applied exploratory factor analysis (EFA) to discover the unobserved latent traits (factors) among pattern-clusters. CT images at total lung capacity (TLC) and residual volume (RV) of 541 former smokers and 59 healthy non-smokers from the cohort of the SubPopulations and Intermediate Outcome Measures in the COPD Study (SPIROMICS) were analyzed. TLC and RV images were registered to calculate the Jacobian (determinant) values for all the voxels in TLC images. 3D Regions of interest (ROIs) with two data channels of CT intensity and Jacobian value were randomly extracted from training images and were fed to the 3D CAE-FC model. 80 pattern-clusters and 7 factors were identified. Factor scores computed for individual subjects were able to predict spirometry-measured pulmonary functions. Two factors which correlated with various emphysema subtypes, parametric response mapping (PRM) metrics, airway variants, and airway tree to lung volume ratio were discriminants of patients across all severity stages. Our findings suggest the potential of developing factor-based surrogate markers for new COPD phenotypes.

Conflict of interest statement

The authors declare no competing interests.

Figures

Figure 1
Figure 1
(a) 3D ROIs are randomly extracted from the 3D CT lung image and fed into the 3D convolutional autoencoder (CAE) to learn the 1D representations (embedding) of the ROIs. A feature constructor (FC) uses the embedding to generate pattern-clusters. (b) The CAE-FC model is applied to the ROIs extracted by the sliding-window technique to classify lung tissue patterns and construct a pattern-cluster-histogram for each subject. Exploratory factor analysis (EFA) is then used to extract the latent traits from the pattern-cluster-histograms of the subjects.
Figure 2
Figure 2
Bar charts showing (a) the averaged intensity of ROIs with large contribution to each factor and (b) the averaged Jacobian of ROIs with large contribution to each factor. Significant differences (p < 0.001) between factors were found in both the averaged intensity and the averaged Jacobian by Welch’s ANOVA. The pairwise Games–Howell post-hoc tests showed that all the pairs were significantly different except for the pair of F5 and F6 that was not significantly different for either the averaged intensity (p = 0.718) or the averaged Jacobian (p = 0.118). (c) Samples of the ROIs taken from the pattern-clusters with strong contribution to each factor.
Figure 3
Figure 3
Factor-representative subjects (i.e. 99th percentile of the factor score among the subjects), shown in a coronal view (posterior–anterior). The pattern-clusters with strong contribution to each factor, ranging from F0 to F6, are shaded with red on the images in the left column. TLC images with CT intensities are displayed in the middle. Jacobian images are shown in the right column.
Figure 4
Figure 4
(a) The correlations between the factors and the emphysema subtypes. The subtypes are CLE, PLE, and PSE. (b) The correlations between the factors and the airway variants. The three variants are: (1) absence of a right medial-basal RB7 branch in RLL (Absent RB7), (2) presence of an accessory sub-superior RB branch in the RLL (Accessory RB), and (3) presence of an accessory LB7 branch in the LLL (Present LB7; see Figure E3). Only correlations with p value less than 0.05 are shown in the figure.
Figure 5
Figure 5
Correlations between each of the seven factors and related clinical data, imaging-based variables, and use of drugs. Only correlations with p value less than 0.05 are shown in the figure.
Figure 6
Figure 6
(a) The original TLC image (left) and the color-coded image (right) of a healthy non-smoker (FEV1/FVC = 79.02%, FEV1% predicted = 81.47%) whose factor scores of F0 and F4 are the closest to the means of the healthy non-smoker cohort. (b) The original TLC image (left) and the color-coded image (right) of a GOLD 0 subject (FEV1/FVC = 56%, FEV1% predicted = 97.88%) whose factor scores of F0 and F4 are the closest to the means of the GOLD 0 subjects. (c) The original TLC image (left) and the color-coded image (right) of a GOLD 4 subject (FEV1/FVC = 29.21%, FEV1% predicted = 20.05%) whose factor scores of F0 and F4 are the closest to the means of the GOLD 4 subjects. The images are plotted in a posterior–anterior view. (d) The averaged factor scores of F0 and F4 for the healthy non-smokers, and GOLD 0–GOLD 4 subjects. (e) The variations of the factor scores for the GOLD 0 subjects are the largest among all subgroups, implying the heterogeneous nature of the subjects at risk.
Figure 7
Figure 7
(a) A confusion matrix of the logistic regression model fitted for the incident of exacerbation, showing the numbers of correctly and wrongly predicted cases. (b) The ROC curve for the logistic regression model. (c) Subjects who experienced exacerbation were significantly different than those who did not in terms of F0.

References

    1. Xu, J., Murphy, S. L., Kochanek, K. D., Bastian, B. & Arias, E. National Vital Statistics Reports Volume 67, Number 5 July 26, 2018, Deaths: Final Data for 2016. (2018).
    1. Kochanek, K. D., Murphy, S. L., Xu, J. & Arias, E. National Vital Statistics Reports Volume 68, Number 9 June 24, 2019 Deaths: Final Data for 2017. (2019).
    1. Agusti A, et al. Characterisation of COPD heterogeneity in the ECLIPSE cohort. Respir. Res. 2010;11:122. doi: 10.1186/1465-9921-11-122.
    1. Galbán CJ, et al. Computed tomography-based biomarker provides unique signature for diagnosis of COPD phenotypes and disease progression. Nat. Med. 2012;18:1711–1715. doi: 10.1038/nm.2971.
    1. Bhatt SP, et al. Association between functional small airway disease and FEV1 decline in chronic obstructive pulmonary disease. Am. J. Respir. Crit. Care Med. 2016;194:178–184. doi: 10.1164/rccm.201511-2219OC.
    1. Hoff BA, et al. CT-based local distribution metric improves characterization of COPD. Sci. Rep. 2017;7:1–12. doi: 10.1038/s41598-017-02871-1.
    1. Haghighi, B. et al. Imaging-based clusters in current smokers of the COPD cohort associate with clinical characteristics: The SubPopulations and Intermediate Outcome Measures in COPD Study (SPIROMICS) 11 Medical and Health Sciences 1102 Cardiorespiratory Medicine and Haemato. Respir. Res.19 (2018).
    1. Haghighi, B. et al. Imaging-based clusters in former smokers of the COPD cohort associate with clinical characteristics: The SubPopulations and intermediate outcome measures in COPD study (SPIROMICS). Respir. Res.20 (2019).
    1. Wedzicha JA. The heterogeneity of chronic obstructive pulmonary disease. Thorax. 2000;55:631–632. doi: 10.1136/thorax.55.8.631.
    1. Schabdach, J., Wells, W. M., Cho, M. & Batmanghelich, K. N. A Likelihood-free approach for characterizing heterogeneous diseases in large-scale studies. in International Conference on Information Processing in Medical Imaging 170–183 (2017). 10.1007/978-3-319-59050-9_14.
    1. Yang, J. et al. Explaining radiological emphysema subtypes with unsupervised texture prototypes: MESA COPD study. in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) vol. 10081 LNCS, 69–80 (Springer, 2017).
    1. Yang, J. et al. Unsupervised discovery of spatially-informed lung texture patterns for pulmonary emphysema: The MESA COPD study. in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) vol. 10433 LNCS, 116–124 (Springer, 2017).
    1. Humphries SM, et al. Deep learning enables automatic classification of emphysema pattern at CT. Radiology. 2019 doi: 10.1148/radiol.2019191022.
    1. Singla, S. et al. Subject2Vec: Generative-discriminative approach from a set of image patches to a vector. in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) vol. 11070 LNCS, 502–510 (Springer, 2018).
    1. Gonzalez G, et al. Disease staging and prognosis in smokers using deep learning in chest computed tomography. Am. J. Respir. Crit. Care Med. 2018;197:193–203. doi: 10.1164/rccm.201705-0860OC.
    1. The National Academies of Sciences, Engineering, Medicine. Chronic multisymptom illness in Gulf War Veterans (2014). 10.17226/18623.
    1. Global Initiative for Chronic Obstructive Lung Disease Pocket guide to COPD diagnosis, management and prevention: A guide for health care professionals. Glob. Initiat. Chron. Obstr. Lung Dis. Inc. 2018 doi: 10.1164/rccm.201701-0218PP.
    1. Cohen J. A power primer. Psychol. Bull. 1992;112:155–159. doi: 10.1037/0033-2909.112.1.155.
    1. Cohen J. Statistical Power Analysis for the Behavioral Sciences. London: Academic Press; 2013.
    1. Barnes PJ. Sex differences in chronic obstructive pulmonary disease mechanisms. Am. J. Respir. Crit. Care Med. 2016;193:813–814. doi: 10.1164/rccm.201512-2379ED.
    1. Kim V, et al. Airway wall thickness is increased in COPD patients with bronchodilator responsiveness. Respir. Res. 2014;15:84. doi: 10.1186/s12931-014-0084-3.
    1. Quanjer PH, et al. Bronchodilator response in FVC is larger and more relevant than in FEV1 in severe airflow obstruction. Chest. 2017;151:1088–1098. doi: 10.1016/j.chest.2016.12.017.
    1. Vestbo J, et al. Changes in forced expiratory volume in 1 second over time in COPD. N. Engl. J. Med. 2011;365:1184–1192. doi: 10.1056/NEJMoa1105482.
    1. Tashkin DP. A review of nebulized drug delivery in COPD. Int. J. COPD. 2016;11:2585–2596. doi: 10.2147/COPD.S114034.
    1. Ikeda A, Nishimura K, Koyama H, Izumi T. Bronchodilating effects of combined therapy with clinical dosages of ipratropium bromide and salbutamol for stable COPD: Comparison with ipratropium bromide alone. Chest. 1995;107:401–405. doi: 10.1378/chest.107.2.401.
    1. Smith BM, et al. Association of dysanapsis with chronic obstructive pulmonary disease among older adults. JAMA J. Am. Med. Assoc. 2020;323:2268–2280. doi: 10.1001/jama.2020.6918.
    1. Paulin E, et al. Influence of diaphragmatic mobility on exercise tolerance and dyspnea in patients with COPD. Respir. Med. 2007;101:2113–2118. doi: 10.1016/j.rmed.2007.05.024.
    1. Tai Joon A, Wan S, Chan Kwon P, Hyoung Kyu Y. The better explanation of COPD: The clinical role of FVC grouping. Eur. Respir. J. 2019;54:PA2640.
    1. Whittaker H, Pimenta J, Kiddle S, Quint J. Rate of FVC decline in a primary care UK chronic obstructive pulmonary disease (COPD) population. Eur. Respir. J. 2019;54:OA1588.
    1. Arjomandi M, et al. Radiographic lung volumes predict progression to COPD in smokers with preserved spirometry in SPIROMICS. Eur. Respir. J. 2019;54:1802214. doi: 10.1183/13993003.02214-2018.
    1. MacNee W, Murchison JT. Small airway disease or emphysema: Which is more important in lung function and FEV1 decline? An old story with a new twist. Am. J. Respir. Crit. Care Med. 2016;194:129–130. doi: 10.1164/rccm.201602-0197ED.
    1. Takahashi M, et al. Imaging of pulmonary emphysema: A pictorial review. Int. J. COPD. 2008;3:193–204. doi: 10.2147/COPD.S2639.
    1. Lynch DA, et al. CT-definable subtypes of chronic obstructive pulmonary disease: A statement of the Fleischner Society1. Radiology. 2015;277:192–205. doi: 10.1148/radiol.2015141579.
    1. Jennermann T, et al. CT-based visual classification of emphysema: Association with mortality in the COPDGene study. Radiology. 2018;288:859–866. doi: 10.1148/radiol.2018172294.
    1. Ballarin A, et al. Mast cell infiltration discriminates between histopathological phenotypes of chronic obstructive pulmonary disease. Am. J. Respir. Crit. Care Med. 2012;186:233–239. doi: 10.1164/rccm.201112-2142OC.
    1. Smith BM, et al. Human airway branch variation and chronic obstructive pulmonary disease. Proc. Natl. Acad. Sci. U.S.A. 2018;115:E974–E981. doi: 10.1073/pnas.1715564115.
    1. Guerra B, Gaveikaite V, Bianchi C, Puhan MA. Prediction models for exacerbations in patients with COPD. Eur. Respir. Rev. 2017;26:160061. doi: 10.1183/16000617.0061-2016.
    1. Couper D, et al. Design of the subpopulations and intermediate outcomes in COPD study (SPIROMICS) Thorax. 2014;69:491–494. doi: 10.1136/thoraxjnl-2013-203897.
    1. Sieren JP, et al. SPIROMICS protocol for multicenter quantitative computed tomography to phenotype the lungs. Am. J. Respir. Crit. Care Med. 2016;194:794–806. doi: 10.1164/rccm.201506-1208PP.
    1. Haghighi BD, Ellingwood N, Yin Y, Hoffman EA, Lin CL. A GPU-based symmetric non-rigid image registration method in human lung. Med. Biol. Eng. Comput. 2018;56:355–371. doi: 10.1007/s11517-017-1690-2.
    1. Yin Y, Hoffman EA, Lin CL. Mass preserving nonrigid registration of CT lung images using cubic B-spline. Med. Phys. 2009;36:4213–4222. doi: 10.1118/1.3193526.
    1. Choi S, Hoffman EA, Wenzel SE, Castro M, Lin CL. Improved CT-based estimate of pulmonary gas trapping accounting for scanner and lung-volume variations in a multicenter asthmatic study. J. Appl. Physiol. 2014;117:593–603. doi: 10.1152/japplphysiol.00280.2014.
    1. Kim SS, et al. Improved correlation between CT emphysema quantification and pulmonary function test by density correction of volumetric CT data based on air and aortic density. Eur. J. Radiol. 2014;83:57–63. doi: 10.1016/j.ejrad.2012.02.021.
    1. Webb WR. Thin-section CT of the secondary pulmonary lobule: Anatomy and the image. Radiology. 2006;239:322–338. doi: 10.1148/radiol.2392041968.
    1. Horn J. Factors in factor analysis. Psychometrika. 1965;30:179–185. doi: 10.1007/BF02289447.
    1. Marsh S, et al. Physiological associations of computerized tomography lung density: A factor analysis. Int. J. Chron. Obstruct. Pulmon. Dis. 2006;1:181–187.
    1. Costello AB, Osborne JW. Best practices in exploratory factor analysis: Four recommendations for getting the most from your analysis. Pract. Assess. Res. Eval. 2005;10:1–9.
    1. Jahani N, et al. Assessment of regional ventilation and deformation using 4D-CT imaging for healthy human lungs during tidal breathing. J. Appl. Physiol. 2015;119:1064–1074. doi: 10.1152/japplphysiol.00339.2015.
    1. Ostridge K, et al. Relationship between pulmonary matrix metalloproteinases and quantitative CT markers of small airways disease and emphysema in COPD. Thorax. 2016;71:126–132. doi: 10.1136/thoraxjnl-2015-207428.
    1. Pope PT, Webster JT. The use of an F-statistic in stepwise regression procedures. Technometrics. 1972;14:327.
    1. Tanabe N, et al. Associations of airway tree to lung volume ratio on computed tomography with lung function and symptoms in chronic obstructive pulmonary disease. Respir. Res. 2019;20:1–8. doi: 10.1186/s12931-019-1047-5.

Source: PubMed

赞助者和合作者

医疗条件

药物干预

3
订阅