Brain Tumor Image Segmentation Method Based on M-Unet Network

Brain tumor image segmentation is a challenging problem in medical image processing. The purpose is to obtain tumor location, size, structure and other information through intelligent processing of medical image data. Accurate brain tumor segmentation algorithm plays an important role in the early diagnosis and treatment planning of tumors. Therefore, this paper proposes a brain tumor segmentation algorithm based on M-Unet network(Multi-scale Unet). This method adds multi-scale convolution modules to the feature fusion part of the original Unet network, so that the network can better extract the multiscale features of the image. At the same time, the cosine annealing learning rate attenuation strategy is used in the training process, so that the network can jump out of the local optimal solution to obtain the best network weight. The results show that the M - Unet network achieves high segment results in MRI image dataset, and the Jaccard coefficient and Dice coefficient of the algorithm reach 0.783 and 0.873 respectively; Compared with the original Unet network, the Jaccard coefficient and Dice coefficient are increased by 2.5% and 4.3% respectively, which shows that M-Unet network can effectively improve the segmentation performance of brain tumor image.