Early prediction of brain tumor classification using convolution neural networks

Abstract

Automatic brain tumor classification of tissue types plays a significant task in computer-aided medical diagnosis. In recent years, classification of brain tumors types like meningioma (T1), glioma (T2), and pituitary tumor (T3). Convolution Neural Networks (CNN), which trains the image into increasingly sub-dividing as filter blocks for the fine-tune of feature extraction from each sub-region, exhibits excellent results and successfully used for object detection and classification. In this paper, we present an approach to improve tumor detection and classification performance. Initially, the tumor area is clustered with the fuzzy c-means algorithm for discovering the surrounded tumor tissues and also gives important clues for tumor types. Second, the Canny edge detection applied for the tumor region. Third, the spectral residual for saliency map from the tumor region. Finally, we combine all three areas into one representation for CNN training and testing on a large dataset, which gives accuracies of 91.40%. These experimental results show that the proposed method is realistic and useful for the classification of brain tumors types.