Brain Tumor Segmentation Based on Deep Learning

Abstract

Brain tumors develop rapidly and aggressively, causing brain damage and can be life threatening. Determining the extent of the tumor is a major challenge in brain tumor treatment planning and quantitative assessment to ameliorate the quality of life of patients. Magnetic resonance imaging (MRI) is an imaging technique widely used to evaluate these brain tumors, but manual segmentation prevented by the large amount of data generated by the MRI is a very long task and the performance is highly dependent on operator's experience. In this context, a reliable automatic segmentation method for segmenting the brain tumor is necessary for effective measurement of the extent of the tumor. There are several image segmentation algorithms, each with its own advantages and limitations. In this paper, we propose a method based on Deep Learning, using deep convolution networks based on the U-Net model. Our method was evaluated on real images provided by Medical Image Computing and Computer-Assisted Interventions BRATS 2017 datasets, which contain both HGG and LGG patients. Based on the experiments, our method can provide a segmentation that is both efficient and robust compared to the manually delineated ground truth. Our model showed a maximum Dice Similarity Coefficient metric of 0.81805 and 0.8103 for the dataset used.