Evidence Fusion with Contextual Discounting for Multi-modality Medical Image Segmentation

As information sources are usually imperfect, it is necessary to take into account their reliability in multi-source fusion tasks. In this paper, we propose a new deep framework allowing us merge multi-MR image segmentation results using the formalism of Dempster-Shafer theory while taking different modalities relative classes. The composed an encoder-decoder feature extraction module, evidential module that computes belief function at each voxel for modality, and multi-modality evidence which assigns vector discount rates modality combines discounted Dempster’s rule. whole trained by minimizing loss based on Dice index increase accuracy reliability. method was evaluated BraTs 2021 database 1251 patients with brain tumors. Quantitative qualitative show our outperforms state art, implements effective idea merging multi-information within neural networks.