Brain Tumor Image Segmentation Based on Capsule Network
Abstract
The overall layered structure, principles and functions of the capsule network and the generation of the confrontation network are introduced. Then the capsule network is applied to discriminator of generation adversarial network, thus the multi-scale generation adversarial capsule network MS-CapsNetGAN model is proposed. The MS-CapsNetGAN model is validated on the MNIST and CIFAR-10 datasets, and the MS-CapsNetGAN model is compared with three mainstream classifiers (Fisherfaces, LeNet and ResNet). The results show that the model has a good segmentation effect. Finally, the MS-CapsNetGAN model performs segmentation experiments on brain MRI images of 100 patients in a hospital, and the Dice coefficient is as high as 93.61%. The results show that the MS-CapsNetGAN model can accurately segment MRI images.