Electronic Thesis and Dissertation Repository

Thesis Format

Integrated Article

Degree

Doctor of Philosophy

Program

Medical Biophysics

Supervisor

Parraga, Grace

Abstract

Pulmonary imaging using computed tomography (CT) and magnetic resonance imaging (MRI) provide a method to measure airway and parenchymal pathologic information that cannot be provided using spirometry. Currently, it remains difficult to predict which chronic obstructive pulmonary disease (COPD) patients will worsen using spirometry, which although safe and inexpensive, does not provide small airway information where COPD is believed to initiate. Quantitative CT and MRI measurements provide regional structure and function information but are not included in mortality risk assessments, prognosis, or COPD staging. Therefore, my overarching hypothesis is that CT and MRI ventilation texture measurements combined with machine learning will classify at-risk ex-smokers, as well as predict accelerated lung function decline and mortality in ex-smokers with and without COPD. I first accurately detected the presence of abnormal diffusing capacity in ex-smokers without COPD or CT evidence of emphysema, by quantifying visually unapparent CT textures and applying machine-learning models. Next, using baseline MR imaging textures, I evaluated longitudinal data to predict accelerated lung function decline in ex-smokers across 3-years. I identified a subset of MRI texture features that independently predicted rapid worsening, where the longitudinal changes of these texture features correlated with changes in lung function. Finally, I used baseline CT and MRI texture measurements and accurately predicted 10-year mortality, which is the ultimate patient outcome. The series of studies presented here are among the first to demonstrate the feasibility of predicting clinically-relevant outcomes exclusively using CT and MR imaging textures. In addition, machine-learning models trained on established clinical and demographic measurements were outperformed by models trained only using texture features. Taken together, these results suggest that quantitative imaging measurements provide additional prognostic value and perhaps should be considered as potential biomarkers for early detection of COPD and evaluating disease progression and longitudinal patient outcomes.

Summary for Lay Audience

It is hard to predict how chronic obstructive pulmonary disease (COPD) will progress and ultimately affect patients in the long run. The limitation arises from the fact that conventional clinical metrics, such as spirometry, offer a global perspective on lung function and may not adequately capture nuances in disease progression. Similarly, established clinical risk factors in such assessments may often capture extra-pulmonary manifestations of the lung disease. The small airways are considered the major site of airflow limitation in COPD, which spirometry measured at the mouth is not sensitive to. In contrast, quantitative computed tomography (CT) measurements allow for the evaluation of airway structural changes, while magnetic resonance imaging (MRI) can provide complementary information on regional ventilation within the lungs. Unfortunately, measurements obtained from chest imaging modalities are currently not included in widely accepted clinical assessments, diagnosis, prognosis, or staging of COPD. Consequently, to address this gap, I developed texture analysis and machine learning algorithms to predict longitudinal clinical outcomes and quantify structural and functional changes occurring in the lungs of ex-smokers with and without COPD. First, I demonstrated the sensitivity of CT texture measurements by detecting abnormal diffusing capacity in patients with clinically-normal CT images. Next, I evaluated MRI texture features at baseline and 3-year follow-up to predict an accelerated lung function decline in ex-smokers with and without COPD. Notably, changes in select texture features over time also correlated with changes in lung function, emphasizing the sensitivity of texture features. Lastly, I predicted 10-year mortality using CT and MRI textures, outperforming all other clinical measurements available to physicians. Together, these results tell us that quantitative imaging textures provide additional prognostic value and perhaps should be considered for evaluating disease progression and clinical outcomes in COPD.

Creative Commons License

Creative Commons Attribution-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-No Derivative Works 4.0 License.

Download

Share

COinS