Improving the Reliability of Lung Densitometry in CT Studies

Abstract

Computed tomography (CT) lung densitometry—namely measurement of lung density—enables quantitative assessment of lung parenchyma and has emerged as a pivotal tool for diagnosing and monitoring of pulmonary diseases such as emphysema. These measurements offer a robust alternative to visual assessment by providing objective, automated, and continuous quantification with shorter evaluation times and better correlation with histopathological measures of emphysema, pulmonary function tests and quality of life measurements. Furthermore, it has shown correlation with pulmonary function tests and quality of life measurements. In recent years, lung densitometry metrics have gained increasing traction in clinical trials for tracking disease progression, owing to their superior sensitivity compared to traditional physiological and health status indices.Reliable and accurate measurement of lung densitometry is essential for its effective use in clinical practice. However, the reliability of these measurements is often compromised by variations in patient-based factors, such as breath-hold volume, as well as differences in scanner hardware and imaging settings. This issue has become increasingly critical with the clinical introduction of photon-counting CT (PCCT), a promising imaging technology that offers superior spatial resolution, contrast, and noise characteristics compared to conventional energy-integrating CT (EICT). This dissertation investigates the influence of variations in scanner technology and imaging settings through a series of experimental studies, identifies imaging parameters that influence lung densitometry reproducibility, and develops methods to mitigate variability and enable harmonized quantification across platforms. Chapters 2-4 investigate the reproducibility of lung densitometry across conventional and emerging CT technologies through three complementary studies: a clinical study, a phantom-based study, and a virtual imaging study. The clinical study compares PCCT and EICT scanner technologies and the influence of imaging parameters in achieving optimum reproducibility. This study further relates observed differences in reproducibility to underlying differences in image quality metrics across scanner technologies. The phantom-based study performs a multi-institutional evaluation of PCCT and EICT scanners across four clinical sites in the United States. This study establishes baseline reproducibility metrics both within and between scanner technologies, quantifies measurement accuracy across a range of acquisition settings, and identifies reproducible protocol pairs across scanner technologies. Finally, the virtual imaging study leverages anatomically realistic virtual human models and scanner-specific CT simulators to compare the accuracy of lung densitometry between PCCT and EICT across diverse imaging conditions. Virtual imaging trials (VITs) provide a controlled and repeatable environment for evaluating CT performance, thus overcoming limitations such as lack of anatomical complexity in physical phantoms or lack of known reference standards and involvement of logistical and ethical constraints in clinical studies. Chapters 5-7 develop methods to mitigate the lack of reproducibility across scanner models and their imaging parameters. Chapter 5 presents a harmonization methodology that transforms CT images into a reference image quality index– defined as spatial resolution and noise index–and lung volume. This framework provides more consistent lung density histograms across imaging conditions. Chapter 6 introduces a deep learning-based methodology for emphysema segmentation that demonstrates greater robustness to variations in imaging parameters compared to traditional lung density biomarkers. This framework enables more accurate emphysema quantification from CT scans and demonstrates improved correlation with radiologist visual assessments and pulmonary function tests. Chapter 7 develops the first set of longitudinal virtual human models with improved realism for lung density studies. The data synthesized using the developed human models and the in-house scanner-specific CT simulator were used for evaluation of post-processing algorithms developed in Chapters 5 and 6. In conclusion, this dissertation advances retrospective and prospective approaches for standardizing lung densitometry. These standardization approaches address a critical need for robustness in lung density measurements, especially given the integration of advanced CT technologies (such as PCCT) into clinical practice. Ultimately, a more robust CT biomarker could increase the sensitivity for longitudinal changes, potentially allowing for higher statistical power and shorter clinical trial duration.