In this paper, we propose a vessel segmentation method in brain CT angiography. Its main idea is to remove bones by a bone-to-bone elimination procedure between two series of images, instead of a time-consuming registration algorithm. Some vessel pixels may be removed by mistake during this bone removal procedure. We invent the idea of conditional dilation to restore lost vessel pixels. Whenever a vessel region is identified, we change its outer boundary to background values so that the later region growing can find out the whole vessel structure. Since the registration computation is no longer required, the segmentation process takes less time and makes it feasible for an interactive 3D segmentation in clinical applications.