Automatic segmentation of low-grade glioma in MRI image based on UNet++ model
Dan Xu, Xidong Zhou, Xuefen Niu, Junwei Wang
Abstract
Abstract Glioma is one of the common brain tumors, and the prognosis of patients with low-grade glioma is relatively good. Therefore, this paper takes the segmentation of low-grade glioma as the research direction, and proposes an automatic segmentation algorithm of low-grade glioma MRI image based on UNet++. Firstly, the sample data in the data set is normalized, and then the sample data is divided into training set, verification set and test set, and data augmentation is performed on the training set; finally, all the data in the training set is used to train the network model. In the training process, in order to alleviate the over-fitting problem of model training, we draw into the dropout after each convolution layer. In order to verify the effectiveness of the proposed model, all intracranial tumor images of patients in the test sample set are selected for segmentation, and the final average Dice coefficient can reach: 89.1%. Compared with the segmentation algorithm based on U-Net, the average Dice coefficient of this algorithm is increased by 3.89%. The experimental results show that this algorithm can achieve better automatic segmentation results for low-level glioma images, provide effective reference for doctors’ diagnosis and surgery, and alleviate the problems of low accuracy and low efficiency caused by manual annotation of images only based on personal experience of doctors.