Distributed Markovian segmentation: Application to MR brain scans

A situated approach to Markovian image segmentation is proposed based on a distributed, decentralized and cooperative strategy for model estimation. According to this approach, the EM-based model estimation is performed locally to cope with spatially varying intensity distributions, as well as non-homogeneities in the appearance of objects. This distributed segmentation is performed under a collaborative and decentralized strategy, to ensure the consistency of segmentation over neighboring zones, and the robustness of model estimation in front of small samples. Specific coordination mechanisms are required to guarantee the proper management of the corresponding processing, which are implemented in the framework of a reactive agent-based architecture. The approach has been experimented on phantoms and real 1.5 T MR brain scans. The reported evaluation results demonstrate that this approach is particularly appropriate in front of complex and spatially variable image models.