Ensemble Transfer Learning for Enhanced Brain Tumor Diagnosis: A new Approach for early detection

Abstract

Brain tumors represent one of the most extreme and complex types of most cancers, requiring unique analysis for powerful remedy and management. Accurate and early identification of brain tumors can greatly enhance patient consequences and decrease mortality. Nowadays deep learning aids the medical field a lot by diagnosing Magnetic Resonance Imaging (MRI) images in Brain tumors. The potential of deep transfer learning architectures to improve brain tumor diagnosis accuracy is explored in this work. This study evaluated three different Convolutional Neural Network (CNN) architectures: AlexNet, VGG16, and ResNet18 as an ensemble model. The gathered dataset was used to train and test the models. In order to increase the dataset's balance and the models' performance, data was collected from: Rizgary Hospital (Erbil), and Hiwa Hospital (slemani). These image enhancement techniques were applied to two categories: normal and abnormal brain tumors. Several brain tumor datasets are available online for the development of Computer Aided Diagnosis systems (CADs), but not KRI Hospital cases, which pose challenges in their classification through deep learning models. This study was implemented by python programming language. Out of the three models used, ResNet had the highest accuracy of 98.66%, VGG16 had an accuracy of 97.8% and AlexNet had an accuracy rate of 97.666%. I also used ensemble between the three models, ensemble predictions of all the models together, majority voting was (98.33%) and the weighting voting was (98.33%).