Brain MRI Segmentation

Our research into the brain segmentation of MRI data has involved reviewing the level set method of surface representation and deformation. Through the efficient narrow band method, which updates scalar field on a number of layers either side of the surface, an automated segmentation method has been developed. A k-means clustering of regularly sampled points has been used to provide an initialisation of seed locations and mean voxel values for the level set framework, removing the requirement of any manual selection of these parameters.

correcting and refining brain segmentation

A problem found with this method, along with other common segmentation methods, was anisotrophy. This causes frequent miscalssifcation of voxels, and clearly results in 3D views of surfaces requiring extensive refinement. Extensions to sparse field and rendering algorithms can efficiently perform localised modification and re-rendering of the surface. This, and the inherent features of a level set approach, allow easy situation of intelligent agent swarms for inhabiting and modifying the zero-level set. Additionally, a simple weighting of movement provides the swarm with direction of user-specified points in order to affect refinement.

Citations: [·]
Last updated 2014 | School of Computer Science