Research On Brain Tumor Segmentation Algorithm Based On Dense And Skipped Connections

Brain tumor is refers to the internal growth of abnormal cells in the brain,because of the complexity of the structure of the brain,the brain tumor image of the lesion area segmentation is very important to the doctor diagnosed accurately,and the depth study of the application in image segmentation tasks more and more widely,under this background,in this paper,the application of deep learning brain tumor image segmentation method is studied,A brain tumor image segmentation method based on dense hopping linkage is proposed.In order to obtain more accurate segmentation results,the analysis and learning of brain tumor image sequence should not only take into account the multi-dimensional spatial information,but also carry out association learning of images under various imaging modes.In order to reduce image noise and increase the number of data sets,the brain tumor images are preprocessed and enhanced.In the design of the network model,in order to better mine the high-dimensional spatial correlation of data,this paper chooses the model based on 3D convolution to complete the segmentation task.In order to make better use of the characteristics of the bottom layer,this paper chooses U-NET as the basic framework,combines the information of the bottom layer and the information of the top layer by means of jump connection,and improves the problem of insufficient information of the upper sampling.In order to improve the reuse rate of features,the Dense module was added to the FRAMEWORK of U-NET to enhance the transmission of features and improve the problem of gradient disappearance in training to some extent.In order to improve the segmentation accuracy of the model,the CBAM attention module is added to make the model pay more attention to the key information in the channel during training.The results show that the improvement of Dense module and CBAM attention module has a positive effect on the model,and the combined model of the two modules has a significant improvement in the segmentation effect.The improvement of ET（Enhance tumor）WT（whole tumor）was 2.60% and 1.06%,respectively.At the same time,in order to verify the effectiveness of the method in this paper,a horizontal comparison experiment was carried out to compare the experimental model in similar studies with the model in this paper.The experimental results show that the model in this paper is superior to other models in the average Dice coefficient performance of the three projects.