Computer Vision News - October 2020

3 Automated Segmentation Of Coronary Arteries 4 “It is now possible to create virtual models of specific patients , and based on this segmentation process, we can convert that to a standard geometry representation that is very easy to work with,” Ramtin explains. “You can plug the virtual models into any research or workflow you are doing. For example, some researchers are looking into how the vessel geometries differ between patients in a large population. The automatic virtual reconstruction can also directly be 3D printed as physical models of patients’ arteries which is very useful for educational purposes or for testing medical devices. The output is a very versatile format and used in a variety of applications.” This is the first challenge to develop fully automatic segmentation methods of the entire coronary artery tree of healthy and diseases patients . Previous challenges have looked at only extracting the centerlines of the vessels, quantifying stenosis, or segmenting specific vessel segments. It has been a large undertaking and in the works for many months, with cardiologists having helped to prepare the manually annotated reference data in time besides major holdbacks due to the global pandemic. The challenge has already wrapped up and results have been collated. The output of the participants’ segmentation algorithm will be compared with the expert annotation, and the 10 highest-performing participants will be presenting their solutions at the event tomorrow. Each team will have a 15-minute presentation on their approach and methodology, followed by questions, and there will be plenty of time for open discussion too. The Automated Segmentation of Coronary Arteries Challenge event will be held at MICCAI on Thursday Oct 8.