Deep Ensemble Feature Extraction based Classification of Brain Tumor using MRI Images

Brain Tumor (BT) classification involving recognition of abnormalities in the brain is challenging since it requires proper imaging techniques such as Magnetic Resonance Imaging (MRI), CT (Computed Tomography), or PET (Positron Emission Tomography) scans. The early detection of the disease is beneficial for its cure. A well-defined methodology for classification of BT is proposed in this paper. Two different dataset containing MR images have been employed; the first one being a combination of two classes named as BT-2 while the other comprises of four classes named as BT-4. A novel hybridized framework incorporating the extraction of features using deep ensemble techniques is proposed. The concatenation of the deep features using VGG 16, ResNet 50 and Inception V3 are implemented into various Machine Learning classifiers for BT classification. Experimental results demonstrate that Support Vector Machine (SVM) generates an impressive average accuracy of 99% using BT-2 dataset and an accuracy of 94% using BT-4 dataset.