Quantification of coronary arterial stenoses in CTA using fuzzy distance transform

Results of curve skeletonization and noise pruning. (a) A region from a binarized coronary artery tree. (b) Curve skeleton of (a) without noise pruning; voxels identified as noise using the proposed post-skeletonization noise pruning algorithm are colored in red. (c) The result after noise pruning.

An illustration of shape distance transform. Both black and textured pixels indicate skeletal pixels in the given shape. Black pixels survive in the skeletonization process as shape pixels, i.e., saved to preserve local shape of the structure. On the other hand textured pixels are preserved to maintain the topology. Shape distance only counts shape voxels on a path. For example, only one shape voxel contributes to the shape length of the path π.

(a and b) Results of expected arterial diameter computation on a healthy arterial branch. The straight line shown in (b) predicts the expected diameter at any location along the arterial branch. (c and d) Same as (a) and (b) but for a branch with a visible stenosis. It is notable in (d) that the method successfully eliminates the artifacts in observed diameters around the stenosis.

Illustration of stenosis simulation. (a) A simulated 3D coronary artery tree and labeling of different branches – right coronary tree: right coronary artery (RCA) and acute marginal (AM); left coronary tree: left main coronary artery (LM), proximal left anterior descending artery (LAD), first diagonal (FD), second diagonal (SD) and proximal left circumflex artery (LC). (b) An example of simulated stenoses on the original phantom data of (a). Locations and severity of stenoses are marked. (c) A maximum intensity projection (MIP) of the image of (b) in the presence of partial voluming and noise.

Results of FDT-based computerized coronary arterial stenosis quantification via CTA. (a) A slice image from a CTA image acquired at 0.43×0.43×1.25mm³ resolution along with boundaries of segmented coronary arteries. (b) 3D rendition of the left coronary artery tree derived from CTA. (c) The medial axis representation of the coronary artery tree after applying skeletal pruning. (d) A color-coded display of observed local diameter in 3D coronary artery tree; the color scale is shown. (e) FDT-based detection and quantification of stenoses. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Illustration of the graphical user interface supported by the CardIQ software for detecting and grading coronary arterial stenoses. (a) 3D reconstruction of heart and coronary arteries. (b) The graphical view using the curved image reformatting. (b) The lumen view and the measurement of local diameter profile (green) along the arterial central line. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Illustrations of more results of coronary artery stenosis quantification. (a) A slice image from a CTA dataset acquired at 0.43×0.43×1.25mm³ resolution along with boundaries of segmented coronary arteries. (b) Detection and quantification of stenoses. (c and d) Same as (a) and (b) but for CTA of another patient.

An illustrative comparison of performances by the current FDT- and BDT-based and conventional methods for quantifying coronary arterial stenoses in simulated realistic phantoms. Here, the total height of each bean represents the average percentage error by a specific method at a particular severity of stenosis; the height of each color band in a given bin indicates the percent error committed on a specific arterial branch.