Automated measurement of heterogeneity in CT images of healthy and diseased rat lungs using variogram analysis of an octree decomposition

摘要

Background Assessing heterogeneity in lung images can be an important diagnosis tool. We present a novel and objective method for assessing lung damage in a rat model of emphysema. We combined a three-dimensional (3D) computer graphics method–octree decomposition–with a geostatistics-based approach for assessing spatial relationships–the variogram–to evaluate disease in 3D computed tomography (CT) image volumes. Methods Male, Sprague-Dawley rats were dosed intratracheally with saline (control), or with elastase dissolved in saline to either the whole lung (for mild, global disease) or a single lobe (for severe, local disease). Gated 3D micro-CT images were acquired on the lungs of all rats at end expiration. Images were masked, and octree decomposition was performed on the images to reduce the lungs to homogeneous blocks of 2?×?2?×?2, 4?×?4?×?4, and 8?×?8?×?8 voxels. To focus on lung parenchyma, small blocks were ignored because they primarily defined boundaries and vascular features, and the spatial variance between all pairs of the 8?×?8?×?8 blocks was calculated as the square of the difference of signal intensity. Variograms–graphs of distance vs. variance–were constructed, and results of a least-squares-fit were compared. The robustness of the approach was tested on images prepared with various filtering protocols. Statistical assessment of the similarity of the three control rats was made with a Kruskal-Wallis rank sum test. A Mann-Whitney-Wilcoxon rank sum test was used to measure statistical distinction between individuals. For comparison with the variogram results, the coefficient of variation and the emphysema index were also calculated for all rats. Results Variogram analysis showed that the control rats were statistically indistinct (p?=?0.12), but there were significant differences between control, mild global disease, and severe local disease groups (p? Conclusion These results suggest the octree decomposition and variogram analysis approach may be a rapid, non-subjective, and sensitive imaging-based biomarker for characterizing lung disease.