Aorta segmentation with a 3D level set approach and quantification of aortic calcifications in non-contrast chest CT

Abstract

Automatic aorta segmentation in thoracic computed tomography (CT) scans is important for aortic calcification quantification and to guide the segmentation of other central vessels. We propose an aorta segmentation algorithm consisting of an initial boundary detection step followed by 3D level set segmentation for refinement. Our algorithm exploits aortic cross-sectional circularity: we first detect aorta boundaries with a circular Hough transform on axial slices to detect ascending and descending aorta regions, and we apply the Hough transform on oblique slices to detect the aortic arch. The centers and radii of circles detected by Hough transform are fitted to smooth cubic spline functions using least-squares fitting. From these center and radius spline functions, we reconstruct an initial aorta surface using the Frenet frame. This reconstructed tubular surface is further refined with 3D level set evolutions. The level set framework we employ optimizes a functional that depends on both edge strength and smoothness terms and evolves the surface to the position of nearby edge location corresponding to the aorta wall. After aorta segmentation, we first detect the aortic calcifications with thresholding applied to the segmented aorta region. We then filter out the false positive regions due to nearby high intensity structures. We tested the algorithm on 45 CT scans and obtained a closest point mean error of 0.52 ± 0.10 mm between the manually and automatically segmented surfaces. The true positive detection rate of calcification algorithm was 0.96 over all CT scans.

Fig. 1

Examples of regions that are true calcifications (b and d) and false positives (a and c) after thresholding step. The algorithm further processes these regions and rejects the false positives.

Fig. 2

Mean ± std of error (top figure), Jaccard coefficient (middle figure) and Dice coefficient (bottom figure) for each data set

Fig. 3

Segmented aorta in 3D for two sample CT scans. Colormap indicates the closest point euclidean distance (in mm) between automatically segmented and manually labelled aorta surfaces (Better viewed in color)

Fig. 4

Calcifications detected manually and by the algorithm are shown with yellow and blue respectively. (Better viewed in color)