Abstract:

Magnetic Resonance Imaging (MRI) is a useful technique for diagnosis of abnormalities that may occur in brain and other tissues. Detection of brain tumors in MRI is a difficult task for physicians. Because they must consider the textural, statistical, morphological and color features of the MR images at the same time to determine the tumors. In this paper, a robust brain MR images classifier based on hybrid features by using Ensemble Neural Network (ENN) is proposed. To increase the robustness of the classifier textural, statistical and color features were used as input to the ENN model. The main advantage of proposed method is reducing image dimension and using ensemble neural networks method to increase accuracy. This ensemble based method combines the base classifiers predictions to increase the proposed model performance and evaluated accuracy and AUC values are 98.70 %, 0.976, respectively. These results were compared with the other methods in literature. 

Keywords: