Brain Tumor Recognition Based on Data Augmentation and Convolutional Neural Network

Abstract

The brain tumor is one of the most dangerous diseases at present. Accurate diagnosis of brain tumors can contribute to improving the prognosis conditions of patients. Existing methods have some shortcomings, such as manual extraction of features and insufficient amount of data. Since the convolutional neural network (CNN) can extract features automatically, we propose a deep Convolutional Neural Network to diagnose the brain tumors. In this paper, an automatic system based on CNN is proposed to classify three categories of brain Magnetic Resonance Images, including normal images, brain images with meningiomas and brain images with gliomas. Several steps, including image preprocessing, data augmentation and image classification, are applied to the original brain images. And the experiments show that the accuracy of the proposed system on testing set can reach 93.33%, which indicates that our model can achieve a comparable classification result.