An Ensemble Learning Approach for Brain Tumor Classification Using MRI

Digital image processing is a prominent tool which is used by radiologists to diagnose the complicated tumor. Magnetic resonance imaging, CT scans, X-rays, etc., are examined and analyzed by extracting the meaningful/accurate information from them. Diagnosing the brain tumor with accuracy is the most critical task. The survival of the infected patients can be increased if the tumor is detected earlier. In this research paper, an ensemble approach is proposed to classify the benign and malignant MRI of the brain. The total of 150 image slices from the Harvard Brain Atlas dataset is utilized in the ratio of 60:40 for training and testing the proposed method. Otsu’s segmentation has been applied to segment the tumor from the skull. Then, the hybrid features including shape, intensity, color and textural features of the MRI are extracted. Decision trees, k-nearest neighbor and support vector machine classifiers are applied separately on the feature set. Then, the stacking model is applied to combine the outcome/prediction of each classifier and gives the final result. The proposed methodology is validated on an open dataset and achieved 97.91% average accuracy, 88.89% precision and 94.44% sensitivity. When compared with other existing methodologies, better accuracy is achieved by this approach.