In recent years, many segmentation methods have been proposed for brain tumor segmentation, among them deeplearning approaches have good performance and have provided better results than other methods. In this paper, an algorithm based on deep neural networks for segmentation of gliomas tumor is presented which is a combination of different Convolutional Neural Network (CNN) architectures. The proposed method uses local and global features of the brain tissue and consists of preprocessing and post-processing steps which leads to better segmentation. The accuracy of the results was evaluated using the dice score coefficient and the sensitivity on the images obtained from two modalities, Flair and T1, from the BraTs2017 data set and achieved acceptable results compared to state of the art methods.
