D30-06 A Subject-Specific Spring Network Model of Human Lung Tissue Predicts Progression in Chronic Obstructive Pulmonary Disease

Abstract Rationale Chronic Obstructive Pulmonary Disease (COPD) is a complex disease that shows significant subject-to-subject variations with multiple stages and phenotypes, such as emphysema, that can vary depending on many factors. Previous studies suggest that enzymatic degradation can lead to a self-sustaining rupture of parenchymal tissue under dynamic mechanical forces of transpulmonary pressure, linking biological action and mechanical failure. Computational modeling can be a powerful tool for investigating interactions between biological activities and mechanical failure during emphysema progression. In the present study, we developed a method for creating subject-specific spring models that allows for mapping computed tomography (CT) images of human lungs onto spring network models. We applied a strain-based mechanism of emphysema progression to these spring networks and compared their predictions to disease progression seen in serial CT images from patients. Methods To model COPD in human lungs, 2D spring networks were made in the transverse plane of National Lung Screening Trial (NLST) CT scans using a novel method that utilizes dithering, cartograms, and stress inference, resulting in a mechanically accurate model of the tissue. A total of 16 slices from 8 patients were used. The corresponding region of tissue one to two years later was identified via image registration. A strain-based disease progression model was then repeatedly applied to the spring network. This caused deformation of the final spring network configuration that was used to deform the original CT image into a prediction image that was compared to the initial and final patient images. The distributions of low attenuation areas (ALA ) within each image were fit with power-laws given by p(ALA)=aALA-b . Results The spring model predicted disease progression consistent with the true progression seen in the human subjects (Fig.1,A-C). Furthermore, the parameters of the power law fit of the cluster size distributions between the initial and final scans were significantly different (p 0.01) and the values between the initial and the simulated scans were significantly different (p 0.01), but the values between the final and simulated scans were not different (Fig. 1 D,E). Conclusions In this study, we developed, for the first time, a method of creating subject-specific spring networks of lung tissue and tested their efficacy in predicting emphysema progression. We found that model predictions were consistent with progression assessed by CT imaging. This represents a substantial step towards using personalized COPD network models as prognostic tools to guide treatment protocols on a personalized level. This abstract is funded by: Provost’s Discretionary Fund