Three-Dimensional Selective-Scale Texture Analysis of Computed Tomography Pulmonary Angiograms

Abstract

Objectives: This feasibility study aims to develop 3-dimensional (3D) selective-scale texture analysis of computed tomography pulmonary angiography to identify texture correlates for ventilated and vascular lung for visual and quantitative assessment of pulmonary disorders with altered vasculature. Materials and Methods: Computed tomography pulmonary angiography examinations of 8 patients were considered in this study; 3 had normal lungs, 3 had pulmonary embolism (PE1, PE2, and PE3), 1 had only emphysema (PEmp), whereas the final patient had both emphysema and embolism (PEE). Before texture analysis, an initial automated segmentation procedure to include only the lung parenchyma and generation of isometric volume were carried out. From this segmented volume, ventilated lung and pulmonary vessels were separately selected. Texture analysis comprised 2 stages: 1) volume filtration using 3D Laplacian of Gaussian filter to highlight fine and coarse textures within ventilated and vascular lung, followed by 2) quantification of texture using mean gray-level intensity, entropy and uniformity both globally and at 3 anatomic sections of the lung, ie, anterior, middle, and posterior. Quantification of texture was also performed on the unfiltered computed tomography lung dataset. Volume rendering and image fusion of ventilated and vascular lung texture were employed for visualization. Results: For fine texture quantified as mean gray-level intensity in ventilated lung, a postural gradient compatible with known pulmonary physiology was demonstrated and texture was different in emphysematous lung (PEmp and PEE) when compared with nonemphysematous lung (normals, PE1, PE2, and PE3) consistent with altered ventilation. Coarse texture in vascular lung demonstrated a descending trend in entropy (or ascending trend in uniformity) for normals, followed by embolism only (PE1, PE2, and PE3) and finally for emphysematous lung (PEmp and PEE) suggesting a correlation with degree of vascularity (or perfusion). 3D images of ventilated and vascular lung texture highlighted mismatched and matched defects in patients with pulmonary disorders. Conclusions: This feasibility study demonstrated that 3D filtered texture analysis can potentially provide correlates for ventilated and vascular lung, which may be useful in the diagnosis of PE in the presence of other causes of altered vascularity.