Brain Tumor MRI Image Segmentation Via Combining Pyramid Convolution and Attention Gate

Gliomas are the most common primary brain malignancies. The precise segmentation of gliomas and their subregions using magnetic resonance images is of great significance to clinical diagnosis and medical research. Due to the morphological variability and location heterogeneity of brain tumors, it is very difficult to automatically segment brain tumor MRI images with high accuracy. In this regard, this paper proposes a brain tumor segmentation algorithm based on the improved U-Net model to achieve accurate segmentation of brain tumors. By adding a pyramid residual module to the encoding end of U-Net, the receptive field and feature extraction capability of the model is increased, and an improved attention gate is used to enhance the feature mapping capability of skip connections. We evaluated the performance of our model on the BraTS 2018 and BraTS 2019 validation datasets, where the Dice scores for the whole tumor, core tumor, and enhancement tumor were 0.885, 0.801, and 0.748, respectively, on the BraTS 2018 validation set. In the validation set of BraTS 2019, the Dice scores of the whole tumor, core tumor, and enhancing tumor were 0.883, 0.758, and 0.684 respectively. The experimental results demonstrate the effectiveness of our model for the task of brain tumor segmentation.