A-UNet: Attention 3D UNet architecture for multiclass segmentation of Brain Tumor

Brain tumor is one of the most deadly disease in this globe. Early detection of brain tumor can increase patient survival. Thus, the segmentation of brain tumor is one of the prerequisite step for detection of brain tumor. Magnetic resonance imaging (MRI) is the gold standard technique for visualization of brain tumor. Four different categories of MRI have been used for visualization of brain tumor. It is a challenging task due to the heterogeneous nature of tissues. A lot of research work has been done in this area using deep learning algorithms, especially in BraTS dataset challenge. BraTS 2020 dataset, having three different category of brain tumors such as Whole Tumor (WT), Tumor Core (TC), and Enhanced tumor (ET), is considered. The reported works on this dataset have achieved less accuracy in segmentation of ET region. In this proposed architecture, with the addition of attention gates in UNet model, the accuracy of ET has been improved. The accuracy of this proposed method is competent with other meth-ods. The results are reported in terms of sensitivity, dice similarity coefficient (DSC), and 95% Hausdorff distance (HD95). The proposed method achieves DSC on BraTS 2020 dataset of 0.92, 0.87, and 0.78 for WT, TC, and ET, respectively.