Research on the automatic division of brain tumors based on improving U-Net3+

At this stage, problems such as incomplete information of the automatic division of brain tumors are still the most difficult problems that need to be solved. This article is an improvement of the U-Net3+model, and the Ma-Unet3+model is proposed. First of all, improve the model on the skeleton, keep the original codec structure, and cut the multi-scale connection. Finally, the improvement model will be experimented on the data sets of braats2018 and braats2019. The experimental results show that the entire tumor DICE coefficient of the improvement model is 91.23%, the DICE coefficient of the tumor core is 87.84%, the DICE coefficient of the tumor is 86.65%. The sensitivity is 92.04%, the sensitivity of the core of the tumor is 88.78%, the sensitivity of the tumor is 84.42%, the distance of the entire tumor is 24.4%. The distance between Dov is 24%. Finally, the superiority and feasibility of the improvement model proved the improvement model with other outstanding models.