Fuzzy reference data estimation in brain tumor segmentation studies

In the present work we propose a novel label fusion strategy specifically oriented to MRI Brain tumor segmentation studies. The salient aspect of the overall strategy is the use of the fuzzy set framework to manage uncertainty in the visual interpretation process. In particular Fuzzy Connectedness principles are used to merge individual labels and generate segmentation representative of a common agreement. Labels are provided by the experts who are asked to manually trace few representative points within the objects of interest. Starting from these multiple seeds the Fuzzy Connectedness algorithm computes the segmentation. The proposed strategy is naturally oriented to integrate uncertain information and then it is expected to manage dissimilarity among input labels. Interaction is drastically limited with respect to a complete manual tracing and the formal fuzzy framework supports in the overall process of estimation without arbitrary solutions. A set of experiments have been conceived and in the context of MRI Brain segmentation studies. Results obtained prove the reliability of the proposed label fusion strategy that can be considered alternative to conventional Voting Rules.