Comparison of brain tumor MRI classification methods using probabilistic features

Brain tumor has remained one of the key causes of death in people of all ages. One way to increase survival rate amongst patients is to correctly diagnose cancer in its early stages. There are several classifiers which can classify cancer images with high accuracy. The goal of this paper was to present a brief survey of the main machine learning methods used in literature to classify brain tumor in MRI images. For an unbiased comparison between the different methods used in literature, gray level co-occurrence matrix probabilistic features(GLCM) were used as input features for training and testing the models. Two methodologies were used to establish the significance of feature reduction in classification accuracy. In the first methodology, the extracted feature set from GLCM was applied to the classifiers for comparison of performance. In the second methodology, principal component analysis (PCA) was used to reduce the extracted features and afterwards the uncorrelated reduced vector was applied to the same classifiers. As a result, it was observed that the reduced uncorrelated features improved the accuracy of all classifiers by 10 to 27%.