This proposal is a competing renewal of a grant on 3D intracerebral vascular imaging. We intend to build upon our earlier work by providing a clinically useful, symbolic description of blood vessels supplying vascular CNS tumors, including both solid tumors and arteriovenous malformations (AVMs). For both the surgeon and the interventionalist, it would be enormously valuable to provide a 3D map that contains the tumor margins, that illustrates 3D parent-child vascular connectivity information, that shows the relationship between arteries and veins in AVMs, and that readily segregates vascular branches supplying a tumor from branches supplying normal brain. These kinds of information are those that the clinican most needs to know. Such information is not directly available to clinicians by any current imaging methodology.
The specific aims of this proposal include: 1) segmentation of solid tumors from MR, 2) segmentation of AVMs from magnetic resonance angiograms (MRA) and MR, 3) registration of MR with MRA data, 4) development of methods of vascular desbription to separate arteries from veins and to delineate which arterial branches (and descendents) supply only tumor and which supply brain, and 5) tests of the accuracy and utility of our methodology. In addition to its clinical goals, this proposal aims to meet a number of challenges in computer vision research, including tumor segmentation using a range of cues, a statistical atlas based definition of named vessels, and automated definition of vessel branchpoints. The methods we propose are unavailable elsewhere and should be profoundly useful to clinicians. We propose a number of studies to evaluate the accuracy and utility of our approach, and we also propose new visualization methods to indicate the uncertainty of our symbolic descriptions of image objects.
