A Novel Framework Called HDU for Segmentation of Brain Tumor

Abstract

It is well known that U-net is the most advanced medical image processing framework, but it performs poorly in processing complex images. DenseNet is a framework for improvement based on U-net, which has been proposed in recent years, with well performance but large parameters compared with U-net. This paper proposes a Half Dense U-net network, which combines the advantages of DenseNet and U-Net, reduces the number of DenseNet parameters and improves the segmentation accuracy. Compared with U-Net, DenseNet and ResNet proposed in recent years, our proposed model can precisely locate the tumor boundary of brain tumors, thus obtaining higher recognition quality.