Effects of Intra-patient Lung Volume Variability on CT-Based Emphysema Quantification: A Virtual Imaging Study

While CT scans can non-invasively diagnose and quantify emphysema, its consistency can be impacted by the variability in the patient's lung volume during the acquisition. Studies to evaluate and mitigate this variability requires imaging of patients under varying inspirations, which is ethically constrained due to possible radiation risk from repeated exposures.

To isolate and quantify the effects of lung volume variability on the consistency of CT-based emphysema measurements using a virtual imaging trial approach.

20 human models with emphysema were created at varying inspirations (70% to 100% of full inspiration) emulating various lung volumes. The models were imaged using a scanner-specific simulator (DukeSim) to generate CT images. Emphysema biomarkers (LAA-950 and Perc15) and global lung density measures were computed from simulated CT images. A linear mixed-effects model was used to analyze the effect of lung volume and total lung volume (TLV) correction methods (physiological and statistical models) on the accuracy of these emphysema measures.

The statistical analysis exhibited a marked influence of inspiration level on both LAA-950 and Perc15 (p-values < 0.001). LAA-950 underestimated the amount of emphysema by 1.44±0.32% (mean ± standard error) for every 1-liter deviation from full inspiration, and the log-log scale analysis revealed that Perc15 underestimated the severity by 1.16±0.02% for every 1% lower lung volume. These deviations were reduced by TLV correction methods.

This study isolated the impact of lung volume variability on the consistency of emphysema quantification by eliminating the confounding factors inherent in clinical studies. Through controlled virtual imaging trial framework, we identified that the physiological model-based TLV correction method was more effective strategy compared to the statistical model for mitigating volume-dependent variability in emphysema assessment. Our study demonstrates an investigation that cannot be readily undertaken with human subjects or simplistic physical phantoms.