User interaction for catheter reconstruction is not a major drawback for prostate catheters which are mainly orthogonal to CT slices and easily identified on a single slice. However, breast catheters can be nearly parallel to CT slices and do not appear simultaneously on a single slice.
To identify them, the operator would have to search for the slice on which each catheter first appears. This is a tedious, time consuming and error prone task.
We present a novel procedure aimed at unsupervised 3D catheter reconstruction form CT images, irrespective of the angle between the catheters and the imaging axis. The proposed procedure is performed in two steps. First, the image processing part, catheters are detected as curvilinear structures on all CT images. The detection is based on modeling the curves (catheters) in an image as a zero crossings of the first directional derivative in the direction of the curve normal. The second step is the 3D reconstruction where detected curves are grouped together to form catheters in 3D space. The curve matching is solved as a combinatorial optimization algorithm.
The algorithm does not rely on CT numbers or on a region of interest, and does not require any user input including the number of catheters. With minor modifications, the algorithm can be used for catheter reconstruction from fluoroscopic images as well. Results are obtained by testing the algorithm on both clinical and simulated data. The obtained reconstructed catheters are compared with manual reconstruction.
This work was supported by MDS Nordion