S. Shit, J. Paetzold, I. Ezhov, A. Sekuboyina, B. Menze clDice - a Novel Connectivity-Preserving Loss Function for Vessel Segmentation IEEE Conference on Computer Vision and Pattern Recognition, Nashville, Tennessee, United States, June 2021. Shit and Paetzold share first authorship. (bib)
	Accurate segmentation of vascular structures is an emerging research topic withrelevance to clinical and biological research. The connectedness of the segmentedvessels is often the most significant property for many applications such as dis-ease modeling for neurodegeneration and stroke. We introduce a novel metricnamelyclDice, which is calculated on the intersection of centerlines and volumesas opposed to the traditional dice, which is calculated on volumes only. Firstly,we tested state-of-the-art vessel segmentation networks using the proposed met-ric as evaluation criteria and show that it captures vascular network propertiessuperior to traditional metrics, such as the dice-coefficient. Secondly, we proposea differentiable form ofclDiceas a loss function for vessel segmentation. Wefind that training onclDiceleads to segmentation with more accurate connectivityinformation, higher graph similarity and often superior volumetric scores.
	This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each authors copyright. In most cases, these works may not be reposted without the explicit permission of the copyright holder.