Aorta and Pulmonary Artery Segmentation in Non-Contrast CT

Accurate measurements of the size and shape of the aorta and pulmonary arteries are important as risk factors for cardiovascular diseases and for Chronic Obstructive Pulmonary Disease (COPD), both of which are leading causes of death worldwide. Dilation of the thoracic aorta (Aortic Aneurysm) entails a high risk of aortic rupture or dissection. Additionally, the ratio of the diameter of the pulmonary artery to the diameter of the aorta is a metric that associates with pulmonary hypertension and was shown to be a strong predictor for exacerbations in patients with Chronic Obstructive Pulmonary Disease (COPD). Non-contrast computed tomography (CT) plays a central role in the imaging of the thoracic aorta and pulmonary arteries. The complete 3D dataset provided by non-contrast CT makes the diameter measurement of these vessels possible. However, performing the measurements manually at different landmark levels is very labor intensive. Automated vessel segmentation and subsequent diameter analysis are therefore desirable. The lack of contrast between blood pool regions and surrounding soft tissue makes automatic vessel segmentation in non-contrast CT scans a challenging task. In this project, an automatic algorithm is developed to extract the ascending aorta, aortic arch and descending aorta, as well as the pulmonary artery trunk, left and right pulmonary arteries, in non-contrast, non-ECG gated chest CT images. After a preprocessing step, multi-scale medialness filtering and path tracking is used to extract the vessel centerlines. A non-uniform dilation of the extracted centerlines is then used to initialize a surface graph optimization algorithm for segmenting the vessels. This automatic approach derives a 3D segmentation of the aorta and pulmonary arteries and a landmark level for the pulmonary artery bifurcation which finally allows measuring cross-sectional areas, diameters, volumes, and shapes of these vessels. The 3D segmentation algorithm performs well with an average Dice similarity coefficient of 0.95 for the aorta and 0.92 for the pulmonary arteries. The correlation between manual and automatic diameter measurements was computed in 13 cross-sectional levels perpendicular to the centerlines at thoracic aorta which resulted in the average inter-class correlation of 0.96. The relation between these measurements and other risk factors and disease will be investigated.

3D Cross section view Automatic aorta (red) and pulmonary arteries (blue) segmentation in 3D view. Right shows few aortic cross-sectional levels and the aortic centerline. Left shows an axial chest CT overlaid with both automatic (red and blue) and manual (green and yellow) segmentations of the aorta and pulmonary arteries, respectively.

Research line

Model-based Medical Image Analysis

Associated people


PhD students