Ensemble of Hybrid CNN Models for Brain Tumor Classification

A brain tumor develops due to the abnormal cell formation within the brain. The greatest challenge before initiating treatment is detecting and classifying tumors from brain MRI pictures since brain tumors have a wide range of appearances and there is a lot of overlap between tumor and normal tissues. In this paper, we propose a novel approach for the diagnosis of brain tumor using an ensemble of three hybrid CNN models. We used three different classifiers with CNN, Support Vector Machine, Random Forest and Multinomial Logistic Regression to classify brain tumors. In order to increase the accuracy, the hyper-parameters of the CNN model are tuned using KerasTuner. The proposed Ensemble Learning approach gives promising results in the classification of brain tumor when evaluated on Figshare brain tumor dataset and achieves an accuracy of 96.08% and thus is a trustworthy alternative to human experts for the brain tumor classification.