Brain Segmentation Based on UNet++ with Weighted Parameters and Convolutional Neural Network
Alin Hou, Lang Wu, Hongjian Sun, Qihao Yang, Hong-Kun Ji, Bo Di Cui, Peng Ji · 2021
The accurate segmentation of brain tumor contour and internal tissue is of great significance to the actual medical treatment. In the multimodal segmentation of brain tumor MRI, the 3D network model is superior to the 2D network model in the learning process. However, in the internal tissue segmentation of brain tumor, the effect is often unsatisfactory. At least at present, there is an urgent need for a method that can accurately segment the internal tissue of brain tumor. In this paper, we optimize the UNet++ network model. The improved UNet++ segmentation network was evaluated on the BraTS 2018 and BraTS 2019 datasets. The average Dice coefficient, Positive Predictive Value (PPV), Sensitivity, Hausdorff Distance and training time of the improved UNet++ network model after weight prediction are 0.87, 0.87, 0.89, 0.83 and 10h respectively.