Coronary artery segmentation and stenosis quantification in CT images with use of a right generalized cylinder model
The proposed method is semi-automatic, as it requires the artery endpoints as input. Its outline is as follows. First, a centerline is extracted between the endpoints, using a modified minimal path approach. Second, 2D contours are extracted in planes orthogonal to the centerline, using a Fast-Marching algorithm with an appropriately tailored speed function and stopping criterion. Third, the contours are used to reconstruct a regularized continuous 3D surface based on a Right Generalized Cylinder model. The contour extraction and the regularization of the model parameters are driven by a Kalman filter. Next, an "idealized" cylinder is reconstructed based on a linear regression of the radii of the most reliable contours. Finally, the stenoses are detected and quantified by comparing the actual and idealized cylinder. The method was evaluated on 24 datasets from the Coronary Artery Algorithm Evaluation Framework. It achieved a moderate segmentation accuracy (Dice similarity index 51%), but a poor stenosis detection rate (sensitivity 34%).
Detection confusion tables
|Calc. cat.||QCA (per segment)||CTA (per lesion)|
The results of this method are based on the following centerlines: own or none.
Detection (QCA per segment / CTA per lesion)
| ||%||rank||%||rank||%||rank||%||rank|| |
|0 (0 - 0.1)||0.0||15.0||0.0||15.0||0.0||18.0||0.0||18.0||16.5|
|1 (0.1 - 10)||0.0||16.0||0.0||16.0||0.0||17.0||0.0||17.0||16.5|
|2 (11 - 100)||14.3||17.0||9.1||18.0||18.2||16.0||6.5||15.0||16.5|
|3 (101 - 400)||20.0||18.0||10.0||18.0||15.8||17.0||6.1||17.0||17.5|
These results are based on 30 datasets and 19 submissions.
Avg. Abs. diff.
| ||%||rank||%||rank||Κ||rank|| |
|0 (0 - 0.1)||53.3||15.0||56.2||15.0||-0.01||14.0||14.5|
|1 (0.1 - 10)||50.7||14.0||53.4||13.0||-0.04||14.0||13.8|
|2 (11 - 100)||52.0||14.0||55.8||13.0||0.01||14.0||13.8|
|3 (101 - 400)||54.0||15.0||59.1||15.0||0.02||13.0||14.0|