A stochastic multi-agent approach for medical-image segmentation: Application to tumor segmentation in brain MR images.

According to functional or anatomical modalities, medical imaging provides a visual representation of complex structures or activities in the human body. One of the most common processing methods applied to those images is segmentation, in which an image is divided into a set of regions of interest. Human anatomical complexity and medical image acquisition artifacts make segmentation of medical images very complex. Thus, several solutions have been proposed to automate image segmentation. However, most existing solutions use prior knowledge and/or require strong interaction with the user. In this paper, we propose a multi-agent approach for the segmentation of 3D medical images. This approach is based on a set of autonomous, interactive agents that use a modified region growing algorithm and cooperate to segment a 3D image. The first organization of agents allows region seed placement and region growing. In a second organization, agent interaction and collaboration allow segmentation refinement by merging the over-segmented regions. Experiments are conducted on magnetic resonance images of healthy and pathological brains. The obtained results are promising and demonstrate the efficiency of our method.