Nils Daniel Forkert1,
Till Illies1, Einar Goebell1, Jens Fiehler1,
Heinz Handels2
1Department
of Diagnostic and Interventional Neuroradiology, University Medical Center
Hamburg-Eppendorf, Hamburg, Germany; 2Institute of Medical Informatics,
University of Luebeck, Luebeck, Schleswig-Holstein, Germany
This work presents a computer-assisted method for the segmentation of the arteriovenous malformation nidus from time-of-flight (TOF) MRA datasets. Therefore, the cerebrovascular system is automatically extracted from the TOF dataset and used for a voxel-wise support vector machine classification into nidus and non-nidus vascular structures based on four features. The resulting classification dataset is used for extracting the nidus using 3D region growing. An evaluation based on fifteen datasets with available manual nidus segmentations from two observers demonstrated that the computer-aided method leads to segmentation results within the range of the inter-observer agreement but with a considerably reduced interaction time.