Feature-enhanced generation and multi-modality fusion based deep neural network for brain tumor segmentation with missing MR modalities

Using multimodal Magnetic Resonance Imaging (MRI) is necessary for accurate brain tumor segmentation. The main problem that not all types of MRIs are always available in clinical exams. Based on the fact there a strong correlation between MR modalities same patient, this work, we propose novel segmentation network case missing one or more modalities. proposed consists three sub-networks: feature-enhanced generator, constraint block and network. generator utilizes to generate 3D image representing modality. can exploit multi-source also constrain synthesize modality which must have coherent with multi-encoder based U-Net achieve final method evaluated BraTS 2018 dataset. Experimental results demonstrate effectiveness achieves average Dice Score 82.9, 74.9 59.1 whole tumor, core enhancing respectively across situations, outperforms best by 3.5%, 17% 18.2%.