Research On Classification And Segmentation Of MRI Of Brain Tumor Based On Deep Learning

Brain tumor is one of the common tumors that threaten human life.Statistics show that China has the highest number of brain tumor incidence and deaths in the world.Brain tumor diagnosis is a complex and difficult task.Magnetic resonance imaging(MRI)is very suitable for the imaging of human brain tissue,and has important significance for the detection and analysis of brain tumors.Doctors can initially judge the existence and specific development of brain tumors by reading MRI images,but the number of MRI images is large and there is interference such as artifacts.Manual judgment is time-consuming and labor-intensive,and there is a risk of missed diagnosis or misdiagnosis.Computer-assisted treatment represented by medical image analysis can help doctors read MRI images.Among them,image classification can help determine different kinds of brain tumor,and image segmentation can help image the shape and location of the tumor.These technologies can improve the diagnosis and treatment of brain tumors,and the rescue of patients with brain tumors.In recent years,traditional image classification and segmentation methods have been gradually replaced by deep learning methods.Deep learning methods do not need to manually extract features like traditional methods,and there are no differences caused by prior knowledge.This paper uses deep learning and computer vision and other related technologies to do brain tumor MRI image classification and segmentation jobs.The specific work is as follows:1.We collect and preprocess the brain tumor MRI image classification and segmentation dataset on the public data platform,then we amplify and adjust the size of dataset to make it fit the model.2.We propose multi-depth concentrated ResNet.Through the analysis of the residual network model and the dataset to prevent the risk of overfitting,this paper proposes a new residual block,and in this new block the convolution layer is modified and the activation function is changed.Experimental results show that the model achieves higher accuracy in classification of brain tumor MRI images.3.We propose a dilated-convolution residual block U-Net.This paper analyzes U-Net model and improves the coding blocks in U-Net.Experimental results and visualizations show that the model can accurately segment MRI images of brain tumors.