Segmentation of Brain Tumors by Evidence Theory: On the use of the conflict information

This paper presents an evidential segmentation scheme of multi-echoes magnetic resonance (MR) images for the detection of brain tumors. The segmentation is based on the modeling of the data by evidence theory which is well suited to represent such uncertain and imprecise data. In our approach, the neighborhood relationship between the voxels are taken into account thanks to a weighted Dempster's combination rule. This process leads to a real region-based segmentation of brain and allows the detection of tumors. In this paper we particularly focus on the conflicting information which is generated when combining neighborhood information. We show this conflict reflects the spatial organization of the data: it is higher at the boundary between the different structures. We propose and define a boundaryindicator based on the amount of conflict. This indicator is then used as new source of evidence that the specialist can aggregate with the segmentation results to soften its decision