Patch and Pixel Based Brain Tumor Segmentation in MRI images using Convolutional Neural Networks

Diagnosing and identifying, as the early-step in the treatment of brain tumors, are of practical importance. Tumors have different shapes, sizes and contrast and appear in any area of the brain. The most common type of tumors is gliomas, which is divided into two categories: low grade glioma and high grade glioma. In this paper, an automated brain tumor segmentation algorithm with a combination of a cascade structure based on convolutional neural network is presented and convolutional neural network. Batch normalization, dropout and nonlinear activation are employed to build architecture. The proposed method uses the BRATS2017 data set. The input images are first divided into patches and then passed through the neural network. Finally, it assigns a label to the central pixel of each patch. The proposed model is evaluated by the standard Dice coefficient and the results are comparable with the state of the art methods.