RELM: A Machine Learning Technique for Brain Tumor Classification

Brain tumor is called as abnormal growth of a group of malicious or benign cells which causes a mass of unwanted cells in the central processing unit of our body, the brain. There are many existing automated techniques to detect them. The detection of these cells is a tedious task and requires great proficiency in order to provide a great deal of cure to them. In the proposed approach, we present an automated brain tumor detection system which not only detects them but also classify them into types based on the features extracted. The brain MRI images needs to be processed and normalized before the system performs the further steps. The pre-processing is key step since it directly affects the quality of classification. Advancing to the next step, the proposed approach uses Principal Component analysis with Normalized GIST (PCA-NGIST) method to extract the features from the brain MRI images. The extracted features, from the datasets are fed to a classifier algorithm for the network to be trained using these features. The training algorithm used in our case is Regularized Extreme Learning Machine (RELM). A test images from the partitioned dataset is given as an input for detection of tumor and further classification if tumor exists. By utilizing the proposed approach, the accuracy percentage of classification rate would be higher.