Latent Correlation Representation Learning for Brain Tumor Segmentation With Missing MRI Modalities

Magnetic Resonance Imaging (MRI) is a widely used imaging technique to assess brain tumor. Accurately segmenting tumor from MR images the key clinical diagnostics and treatment planning. In addition, multi-modal can provide complementary information for accurate segmentation. However, it's common miss some modalities in practice. this paper, we present novel segmentation algorithm with missing modalities. Since it exists strong correlation between multi-modalities, model proposed specially represent latent multi-source correlation. Thanks obtained representation, becomes more robust case of modality. First, individual representation produced by each encoder estimate modality independent parameter. Then, transforms all representations representations. Finally, across are fused via attention mechanism into shared emphasize most important features We evaluate our on BraTS 2018 2019 dataset, outperforms current state-of-the-art methods produces results when one or missing.