More Than Lung Cancer: Automated analysis of low-dose screening CT scans

O.M. Mets

Promotor: M. Prokop and J. W. J. Lammers
Copromotor: P. A. de Jong and P. Zanen
Utrecht University, The Netherlands
October 16, 2012


Lung cancer screening in heavy smokers using low-dose CT has recently been shown to substantially reduce mortality, and practical guidelines on lung cancer screening have been released. This CT-based screening may provide the appropriate setting to thoroughly evaluate a heavily smoking population at high- risk for multiple smoking-induced diseases. However, it was unknown whether low- dose lung cancer screening CT scans enabled additional evaluation of chronic obstructive pulmonary disease (COPD) and cardiovascular risk calculation. If these CT scans would prove suitable for this additional evaluation and the diseases can be treated before becoming clinically apparent, this would reduce the burden of smoking-induced diseases on top of the lung cancer mortality reduction, increasing both the clinical value and the cost-effectiveness of CT- based screening. Therefore, this thesis showed the optimization of automated analysis in expiratory CT and reported the diagnostic and prognostic value of CT for evaluation of diseases other than lung cancer. Part 1 focuses on Automated assessment of air trapping on expiratory CT. We started in Chapter 2 with comparing the variability of CT air trapping measures in heavy smokers on repeat examinations. Our results showed that the expiratory to inspiratory ratio of mean lung density (E/I-ratioMLD) is less sensitive to differences in breath hold reproducibility compared to the percentage of voxels in expiratory CT with an attenuation value below -856 Hounsfield Unit (EXP-856). Also, it does not encounter difficulties when commonly suggested statistical volume correction is applied, contrarily to EXP-856. Chapter 3 expands the current knowledge by reporting the effects of an upcoming and increasingly used iterative reconstruction technique that increases image quality of scans obtained with reduced radiation dose. We showed that this iterative reconstruction technique significantly influences both quantitative CT emphysema and CT air trapping results, however, E/I-ratioMLD was found to be unaffected. In Chapter 4 we compared three different CT air trapping measures suggested in literature 10-12 against a functional reference standard of air trapping in a population of 427 male subjects to identify the optimal CT air trapping measure in a lung cancer screening setting. We showed that E/I-ratioMLD significantly outperforms the other two measures in the detection of air trapping in a lung cancer screening setting. This implicates that when CT air trapping is to be assessed in low-dose screening chest CT, E/I-ratioMLD should be preferred over other quantitative CT air trapping measures. This superiority in air trapping detection is another advantage of E/I-ratioMLD, on top of the robustness shown in the previous chapter. Lastly, in Chapter 5 CT air trapping was found to be independently associated with lung function in a population of COPD subjects, and the combination of CT air trapping and CT emphysema explained around three- quarters of the variability in lung function, compared to about half of the variability by CT emphysema alone. Part 2 focused on lung cancer screening, in particular the automated evaluation of cardiovascular disease and COPD Smoking-induced mortality and morbidity is not only due to lung cancer, but also due to cardiovascular disease (CVD) and chronic obstructive pulmonary disease (COPD). Additional early detection of these smoking-induced diseases in a lung cancer screening setting may reduce morbidity and mortality, and screening for multiple diseases may enhance the cost-effectiveness of CT-based screening in heavy smokers. In Chapter 6 we derivated a prediction model for cardiovascular events in a screening cohort of over 1800 heavily smoking males, and validated the model in another cohort of 1725 screening participants. We found that coronary calcium may lead to an increase of 3.2 times the risk for a cardiovascular event in the follow-up period, and up to 1.9 times the risk for aortic calcium. In Chapter 7 we applied the CT air trapping measures examined in part I of this thesis, and investigated whether an automated model based on inspiratory and expiratory CT densitometry, together with some basic patient characteristics, could accurately diagnose COPD. Our results showed that automated analysis of inspiratory and expiratory CT enables identification of COPD subjects in a screening population with an accuracy of 78 We compared the performance of the automated CT model to that of human observers in Chapter 8. A total of eight observers with various levels of expertise in reading chest CT images were asked to evaluate whether COPD was present or not, based on the paired inspiratory and expiratory CT images and some basic patient characteristics. The performance of the automated CT model was approached only by the specialized chest radiologist. Moreover, all observers showed considerable intraobserver variation on repeat evaluation, regardless of the level of expertise. These results support the potential role of automated evaluation using lung densitometry, as CT images in screening practice may not always be handled by a specialized chest radiologist.

A pdf file of this publication is available for personal use. Enter your e-mail address in the box below and press the button. You will receive an e-mail message with a link to the pdf file.