A Visually Guided Framework for Lung Segmentation and Visualization in Chest CT Images

Abstract

Lung cancer is the leading cause of cancer-related death worldwide and this also stimulates the development of various computeraided diagnosis (CAD) systems. But the conventional lung segmentation methods can't satisfy the needs of the clinicians in lung cancer diagnosis and surgery. It is very important to provide a segmentation and visualization framework for the clinicians instead of radiologists in outpatient service. Therefore we propose a visually guided method based on a 2D feature space and spatial connectivity computation to reduce the dependence on the radiologists for lung segmentation and visualization. Our framework consists of three main processing steps. Firstly, a 2D feature space of CT scalar versus gradient magnitude is constructed. Secondly, the attribute distribution region of the lungs is selected in the 2D feature space, and then the lungs are extracted from the determined voxels by spatial connectivity computation. Finally, the lungs and pulmonary nodules are visualized simultaneously with different colors and opacities in volume rendering. Experimental results show that the proposed framework is efficient for outpatient service and can provide an intuitive segmentation process and nodules information.

COVID-view: Diagnosis of COVID-19 using Chest CT

“…Region-growing [5,9,10] has been proposed for isolating and visualizing the lungs and their internal features. Lan et al [45] have used the selection of voxels over intensity-gradient histograms and spatial connectivity for visualizing lungs and their structures. Automatic semantic labeling of bronchial tree [40,65] can support further analysis and visualization.…”