A major challenge in brain tumor treatment planning and quantitativeevaluation is determination of the tumor extent. The noninvasive magneticresonance imaging (MRI) technique has emerged as a front-line diagnostic toolfor brain tumors without ionizing radiation. Manual segmentation of brain tumorextent from 3D MRI volumes is a very time-consuming task and the performance ishighly relied on operator's experience. In this context, a reliable fullyautomatic segmentation method for the brain tumor segmentation is necessary foran efficient measurement of the tumor extent. In this study, we propose a fullyautomatic method for brain tumor segmentation, which is developed using U-Netbased deep convolutional networks. Our method was evaluated on Multimodal BrainTumor Image Segmentation (BRATS 2015) datasets, which contain 220 high-gradebrain tumor and 54 low-grade tumor cases. Cross-validation has shown that ourmethod can obtain promising segmentation efficiently.
展开▼
