Vascular Atlas Formation Via a vessel-to-image Affine Registration Method Dini Chillet, Stephen Aylward, Julien Jomier Purpose: We have developed a novel method to form a vasculature atlas via distance map. The aim is to be able to detect abnormal vessels, by registering a vasculature from a new subject to our atlas. Abnormal blood vessels, can be in most of the cases, a sign of tumors or aneurysm. The method has been used to form brain and liver normal vasculature atlases. Materials and Methods: The protocol can be divided in to three different steps. First, blood vessels are extracted from each MR/CT scan using a tubular object centerline method. Secondly, the Distance Map Image (DMI) is computed. Then, each vasculature is registered with the DMI using our vessel-to-image affine registration method. The last step is to compute the DMI of the registered vasculatures and average those distance maps. This forms the atlas. Results: We have collected MR and CT angiogram of 10 livers and 13 brains. The set of data comes from healthy volunteers. The cross-patients affine registration shows accurate results. The robustness has been evaluated via Monte Carlo simulation using for the offset a range of +/- 10 voxels and +/- 0.15 radians for the rotation. We are currently forming the atlas, and registering subjects with this atlas and make observations. Conclusions: The registration has shown encouraging results, even though cross-patients registration is hard given big differences in the vasculature within a population. We are testing the efficiency of the comparison between a subject and the atlas. Presented at the 2003 Radiology Research Symposium