(B-49) Mechanical Deformation of the Acutely Injured Lung Using Quantitative CT Imaging

Introduction:: Global mechanical changes of the lung depend on the ensemble behavior of its regional structural elements, which become altered during acute injury. Quantitative computed tomographic (qCT) image analyses have evolved into powerful methodologies for noninvasive assessment of parenchymal mechanics and pathology in inflammatory lung diseases. Segmentation-based qCT estimates of lung parenchymal mechanics can provide global information on lung mechanical deformation during inflation or deflation. However, information on both global and regional lung deformation can be obtained using image registration, a process by which images are aligned by spatial mapping of voxels from one image to that of another. Image registration has become a useful tool for three-dimensional (3D) medical imaging analysis, where estimates of regional parenchymal strain and deformation may be obtained at voxel-scale resolution. Accordingly, image registration may allow for unique insight and characterization of local structure-function alterations associated with lung disease. The purpose of this study was to compare segmentation vs. registration-based qCT techniques to evaluate the mechanical properties of porcine lungs under baseline conditions and following acute injury.


Materials and Methods:: Whole-lung CT scans were obtained in five anesthetized porcine subjects before and after oleic acid-induced lung injury, at constant airway pressures during deflation from 30 to 0 cm H₂O in decrements of 5 cm H₂O. Automatic segmentation of the lung was performed using a convolutional neural network (Gerard et al., 2020). Total volume at each inflation pressure was determined from the sum of voxel volumes within the segmented lung, and segmentation-based distensibility was computed from the change in total volume divided by the change in airway pressure. Following segmentation, pairs of images at adjacent pressures (i.e., 5 to 0 cmH₂O and 10 to 5 cm H₂O) were matched using non-rigid image registration. Regional deformations for each pressure pair were evaluated from the Jacobian matrix of the warping function, which contained the partial derivatives defining its differential deformation. Regional deformation as a scalar value was expressed as its Jacobian determinant |J_n|, with |J_n| ˂ 1 indicating compression, and |J_n| ˃ 1 indicating expansion. The proportions of these regional deformations were determined for all pressure pairs across all subjects, for both baseline and injured conditions. The Jacobian determinants were mapped onto the corresponding CT images, for anatomic visualization of the regional compression and expansion across the entire lung. Voxel-level distensibility in the registered images were summed to yield a registration-based estimate of total lung distensibility, which was then compared to the segmentation-based distensibility.


Results, Conclusions, and Discussions:: Significant correlations were found between registration-based distensibility and segmentation-based distensibility at all inflation pressure pairs, with r² = 0.97, p < 0.01 and r² = 0.98, p < 0.01 for baseline and injured conditions respectively (Fig. 1A&B). More scatter was observed in the correlations at higher values of distensibility. Following lung injury, total lung volume for each airway pressure was unchanged compared to baseline conditions and demonstrated similar relationships between registration-based and segmentation-based methods. Under baseline conditions, lung distensibility generally decreased as inflation pressure increased, and was significantly affected by injury. For lower inflation pressures, incremental distensibility was significantly lower after lung injury compared to baseline (i.e., 10 to 5 cm H₂O and 5 to 0 cm H₂O). However, distensibility was significantly higher following lung injury compared to baseline for the 25 to 20 cm H₂O inflation. For both conditions, the overall increase in inflation pressure from 0 to 30 cm H₂O was followed by an increase in percent regional lung compression (0.16% - 32.1%, and 17.1% - 35.7%), and a decrease in percent regional lung expansion (98.3% - 67.9%, and 82.9% - 64.3%), although the net total lung volume increased. The anatomic patterns of regional compression or expansion were influenced by gravity under baseline or injured conditions, but regional deformations were heterogeneously spread across the lung within each subject.

These data demonstrate that registration-based estimates of lung distensibility were significantly correlated with segmentation-based estimates, and provide insight into the pressure-volume as well as structure-function relationships of the lung in healthy and injured conditions. While segmentation-based methods are restricted to estimating global mechanical changes, image registration Jacobian-based methods provide further insight into regional structure-function derangements associated with acute lung injury, and may serve as a template for characterizing the severity of lung injury in patients.


Acknowledgements (Optional): :

References (Optional): : Gerard, S. E., Herrmann, J., Kaczka, D. W., Musch, G., Fernandez-Bustamante, A., & Reinhardt, J. M. (2020). Multi-resolution convolutional neural networks for fully automated segmentation of acutely injured lungs in multiple species. Medical Image Analysis, 60. https://doi.org/10.1016/j.media.2019.101592