Classification of pulmonary Arteries and Veins (CARVE14)

This page contains data, reference standards, and results from the paper:

IEEE Trans Med Imaging. 2016 Mar;35(3):882-92. doi: 10.1109/TMI.2015.2500279. Epub 2015 Nov 12.

Automatic Pulmonary Artery-Vein Separation and Classification in Computed Tomography Using Tree Partitioning and Peripheral Vessel Matching.

(https://www.ncbi.nlm.nih.gov/pubmed/26584489).
 
The scans used for the evaluation of our method are publically available through the ANODE challenge or via the page CTscans on this website. Part of the results of the method and all manual annotations that were used for the evaluation are available on this page. In this way, we hope to encourage researchers to use the same set of scans and to compare their results with our method. Some results are shown in the figures below, with in figure 1 3D renderings of one of the cases used in our paper, and in figure 2 classification results of four different attached arteries and veins.  

 

Figure 1: 3D renderings of an example case presented in our paper, with in (a) the vessel segmentation, (b) the full reference standard, and in (c) the result from the proposed automatic method. The evaluation of this case resulted in an accuracy of 95%. The white arrows in (a) indicate errors in the vessel segmentation due to diseases in the lungs. These non-vascular structures and their attached subtrees are automatically exclude from the results. Note that this is not affecting the classification of the surrounding vessels.

 
 
 
 
Figure 2: Four example of correctly classified artery-vein attachments by the automatic method. Each of the rows represent a di?fferent artery-vein attachment. In the first frame of each row, an enlarged part of an axial slice of the original scan is presented at a location where an artery and a vein either touch or nearly touch each other. The second frame shows the same area including the results of the vessel segmentation, where the red arrows indicate locations where arteries and veins are attached in the vessel segmentation. The third frame shows the results from the proposed automatic artery-vein classification method, where the blue vessels represent the arteries and the red vessels the veins. A 3D rendering of the same results is shown in the fourth frame, indicating that the method correctly separated the artery-vein attachments.