Graduate project

Automatic Brain Tumor Segmentation with Convolutional Neural Network

There are multiple types of Brain Tumors, which can be difficult to evaluate that leads to unpleasant result for the patient. Thus, detection and treatment planning of the brain tumor is the most important factor in the process. Magnetic resonance imaging (MRI) is broadly used technique to evaluate the brain tumors. Manual segmentation of brain tumor from MRI consumes more time and depended on the experience of the machinist. Thus, automated techniques for the segmentation are required to ease the treatment planning. Even in the automated methods for the segmentation is not so easy because of the various types of the brain tumors. Thus, it is necessary to have reliable method for brain tumor segmentation which can measure the tumors efficiently and less time consuming. In this paper, we propose a technique for brain tumor segmentation which is created using U-Net based convolutional neural network. The technique was evaluated on datasets called Multimodal Brain Tumor Image Segmentation (BRATS 2019). This dataset contains more than 76 cases of low-grade tumor and 259 cases of high-grade tumor.