MIRAU-Net: An improved neural network based on U-Net for gliomas segmentation

Gliomas are the largest prevalent and destructive of brain tumors have crucial parts for diagnosing treating MRI during segmentation using computerized methods. Recently, U-Net architecture has achieved impressive tumor segmentation, but this role remains challenging due to differing severity appearance gliomas. Therefore, we proposed a novel encoder–decoder called Multi Inception Residual Attention (MIRAU-Net) in work. It integrates residual inception modules with attention gates into further enhance performance. Encoder–decoder is connected through pathways decrease distance between their maps features. We use weight cross-entropy generalized Dice (GDL) focal Tversky loss functions resolve class imbalance problem. The evaluation performance MIRAU-Net checked Brats 2019 obtained mean dice similarities 0.885 whole tumor, 0.879 core area, 0.818 enhancement tumor. Experiment results reveal that suggested beats its baselines provides better efficiency than recent techniques segmentation.