Fig. 2:

An overview of the proposed framework is illustrated. First, the MRI images are pre-processed and cleaned using the approaches explained in Section II-B. The cleaned images are then combined to create an input tensor, which is then fed into a seed selection scheme. We have discussed two seeding strategies i.e. affinity propagation (AP) and morphology based operators. The final stage devises two alternative fusion strategies for multi-object segmentation from multi-contrast MRI: fusion at the affinity space and fusion at the decision level (last column). While affinity combination uses evidence based inference from probability density function and segmentation results of each contrast image, decision fusion uses Karnaugh map based multi-label fusion strategy to refine the final tissue segmentation.