A Scale-Space Medialness Transform Based on Boundary Concordance Voting
Journal of Mathematical Imaging and Vision
Model-based detection of tubular structures in 3D images
Computer Vision and Image Understanding
Muliscale Vessel Enhancement Filtering
MICCAI '98 Proceedings of the First International Conference on Medical Image Computing and Computer-Assisted Intervention
Knowledge management in image-based analysis of blood vessel structures
Knowledge and Information Systems
Detection of type II endoleaks in abdominal aortic aneurysms after endovascular repair
Computers in Biology and Medicine
Hi-index | 0.00 |
Accurate detection and extraction of 3D vascular structures is a crucial step for many medical image applications that require vascular analysis. Vessel tracking algorithms iteratively follow vascular branches point by point, obtaining geometric descriptors, such as centerlines and sections of branches, that describe patient-specific vasculature. In order to obtain these descriptors, most approaches use specialized scaled vascular feature detectors. However, these detectors may fail due to the presence of nearby spurious structures, incorrect scale or parameter choice or other undesired effects, obtaining incorrect local sections which may lead to unrecoverable errors during the tracking procedure. We propose to combine this approach with an evolutionary optimization framework that use specific modified vascular detectors as cost functions in order to obtain accurate vascular sections when the direct detection approach fails. We demonstrate the validity of this new approach with experiments using real datasets. We also show that, for a family of medialness functions, the procedure can be performed at fixed small scales which is computationally efficient for local kernel-based estimators.