Train a 3D U-Net to segment cranial vasculature in CTA volume without manual annotation | IEEE Conference Publication | IEEE Xplore