Multi-modal brain tumor image segmentation based on improved U-net model

Brain tumor is a disease that seriously endangers human life. Multi-modal brain tumor magnetic resonance images provide doctors with great help in the timely diagnosis of brain tumor diseases. This paper proposes a multi-modal brain tumor MRI image segmentation method based on the improved U-net model. The model uses dilated convolution to expand the receptive field, obtain more information in the feature, and make up for the deficiencies caused by limited computing resources. At the same time, a global reasoning unit is introduced into the model, which gathers a set of features in the MRI image globally in the coordinate space, and realizes global reasoning in the coordinate-interactive space through the global reasoning pool and weighted broadcasting, which further improves the performance of the backbone architecture. The experiment uses the data set provided by BraTS (The Brain Tumor Image Segmentation Challenge) to verify the improved model and uses the average dice coefficient to evaluate the segmentation results. The preliminary results of mean dice coefficients are 0.89, 0.82, 0.79 for the whole tumor, tumor core, and enhancing tumor respectively.