Segmentation of cerebral vessels and aneurysms from MR angiography data

Abstract

A three-dimensional representation of cerebral vessel morphology is essential for neurologists treating cerebral aneurysms. However, current imaging techniques cannot provide such a representation: slices of MR angiography (MRA) data can only give two-dimensional descriptions, and ambiguities of aneurysm position and size arising in X-ray projection imaging can often be intractable. To overcome these problems, we have established a new, fully automatic, statistically based algorithm for segmenting the three-dimensional vessel information from time of flight (TOF) MRA data. We introduce a mixture distribution for the data, motivated by a physical model of blood flow, that is used in a two stage segmentation algorithm. In the first stage we apply an expectation maximisation (EM) algorithm as a statistical classifier. We then utilise structural criteria in the second stage to refine the initial segmentation. We present results from applying our algorithm to two real data sets, containing both vessel and aneurysm structures.