3D level set image segmentation refined by intelligent agent swarm

Abstract

The level set method of surface representation and deformation has found many applications in image processing, especially with regard to segmentation. Naive numerical solutions have long since given way to much more efficient narrow band methods, where updates to the scalar field are performed only within a number of layers either side of the surface. This paper presents our implementation of automated segmentation via the sparse field narrow band approach. We use k-means clustering of regularly sampled points to provide initialisation of seed locations and mean voxel values, removing the need for manual seed and parameter selection. We demonstrate this method by segmenting grey matter and white matter from a simulated T1 MRI scan.Anisotropy can easily cause this and other segmentation methods to misclassify voxels, however. Often it is clear in a 3D view of the resulting surface which regions require refinement. With available extensions to the sparse field and rendering algorithms we can now efficiently perform purely localised modifications and re-rendering of the surface. These extensions, plus properties inherent in the level set method, allow us to easily situate swarms of agents on the surface, intelligently inhabiting and modifying the zero-level set. By a simple weighting of movement direction, we can direct the swarm to any user-specified point on the surface in order to affect repairs and refinement.978-1-4244-8126-2/10/$26.00 ©2010 IEEE

A survey on image segmentation using metaheuristic-based deformable models: state of the art and critical analysis

“…Moreover, many deformable models are based on partial differential equations which can be solved by traditional numerical methods. However, metaheuristics have demonstrated to be very useful for learning the parameters of the model [196][197][198][199][200], to refine the results obtained by the geometric approach [201], to initialize the contour and/or extract the prior information which is to be used by the level set method [195,198,202] or to directly guide the optimization process avoiding local minima [190][191][192][193]203]. An important advantage of using metaheuristics is that they can optimize the level set function without the need to compute derivatives, thereby permitting a straightforward introduction of new curve-evolution terms [198].…”

“…In [201], agents inhabit the zero-level surface, sensing and modifying it as necessary using straightforward interpolation routines. They are allowed to modify the surface at their location whilst maintaining the structure of the sparse field.…”

“…In [201], each agent is affected by 'forces' pulling it in a certain direction. These forces are weighted and summed and the resulting vector is normalized to compute a movement potential vector, provided by the function P(x).…”

First results and future developments of the MIBISOC Project in the IBISlab of the university of parma

“…In particular, we have relied on the support of members of the School of Computer Science of the University of Nottingham (United Kingdom), experts in the application of geometric deformable models to MRI brain image segmentation [16], in implementing and the Level Set method and applying it to the segmentation of the hippocampus in histological images. As well, we are collaborating with the IRIDIA in the Université Libre de Bruxelles (Belgium) on the automatic parameter configuration [30] of the different GPU implementions of metaheuristics, that were compared in terms of efficiency, parallelism and execution time.…”

Biomedical image segmentation using geometric deformable models and metaheuristics

“…In [30], a GA is used to find an optimal set of parameters that characterize the LS method in CT and MRI segmentation. In [39], the initial segmentation based on the LS method is refined using swarms of intelligent agents. Finally, in [40], a comparative study on the segmentation of histological images is carried out where different geometric approaches are initialized using metaheuristics and parametric DMs.…”