Automatic Brain Tumor Segmentation with Scale Attention Network

Automatic segmentation of brain tumors is an essential but challenging step for extracting quantitative imaging biomarkers accurate tumor detection, diagnosis, prognosis, treatment planning and assessment. Multimodal Brain Tumor Segmentation Challenge 2020 (BraTS 2020) provides a common platform comparing different automatic algorithms on multi-parametric Magnetic Resonance Imaging (mpMRI) in tasks 1) MRI scans; 2) Prediction patient overall survival (OS) from pre-operative 3) Distinction true recurrence related effects 4) Evaluation uncertainty measures segmentation. We participate the image challenge by developing fully network based encoder-decoder architecture. In order to better integrate information across scales, we propose dynamic scale attention mechanism that incorporates low-level details with high-level semantics feature maps at scales. Our framework was trained using 369 training cases provided BraTS 2020, achieved average Dice Similarity Coefficient (DSC) 0.8828, 0.8433 0.8177, as well $$95\%$$ Hausdorff distance (in millimeter) 5.2176, 17.9697 13.4298 166 testing whole tumor, core enhanced respectively, which ranked itself 3rd place among 693 registrations challenge.