Tensor scale-based anisotropic region growing for segmentation of elongated biological structures

Over decades, segmentation has remained a salient task in most medical imaging applications confronting multi-faced challenges including limited image quality. In this paper, we present a new anisotropic region growing segmentation approach for vascular or other elongated structures. A fundamental challenge during tracing vascular structures is broken continuity of structures by noise and other imaging artifacts coupled with leaking through blurring and soft boundaries. Anisotropic region growing solves this problem using tensor scale that captures local structure orientation and geometry using an ellipsoidal model. A new fuzzy connectivity based algorithm is developed that uses tensor scale to facilitate region growing along the local structure while arresting cross-structure leaking. The performance of the method has been quantitatively evaluated on non-contrast human pulmonary CT imaging and the results found are promising.