Segmentation of the pulmonary vascular trees in 3D CT images using variational region-growing

ObjectivesThe long-term goal of this project is to quantify the aeration of lung parenchyma in 3D CT scans of patients with acute respiratory distress syndrome. This task requires lung delineation, as well as elimination of airways and vessels. The objective of this article was to present and evaluate the method used to segment out the vascular trees.Materials and methodsVascular trees are segmented by variational region growing. This process is performed within a lung mask, where the airways and bronchial walls were previously eliminated by adaptive multi-scale morphological operations. The region growth starts from seeds defined as the most salient points on a “vesselness” map. The “vesselness” function based on the eigenvalues of the Hessian matrix is also used in the region descriptor that controls variational region growing. The formulation of this descriptor, as well as the method used to eliminate the bronchial walls, are the original contributions of this work. The method was evaluated using the full set of 20 chest scans from the VESSEL12 challenge framework.ResultsOverall specificity of 0.938 and sensitivity of 0.772 were achieved. The method successfully differentiated vessels from bronchial walls (specificity, 0.848) but failed to detect the smallest vessels (sensitivity, 0.418).ConclusionTo the best of our knowledge, similar formulations of variational region growing have never been used to segment pulmonary vascular trees. The method seems to be suitable for the intended application, although its validation on actual images with acute respiratory distress syndrome remains to be done.