Majority Voting-Based Tumor Detection: Brain Tumor Detection on X-Ray and MRI Images Using Fusion of Hybrid Learning Methods

Abstract

Brain tumor is a crucial health problem that affects people's lives and their life quality. The treatment of this disease varies according to the size, type, location, and condition of the tumor detected. Therefore, the treatment of this disease may vary from person to person. Early diagnosis of brain tumor in individuals is very important as it can change the course of treatment. The aim of this study is to provide a hybrid learning solution for early detection of brain tumors on Magnetic Resonance Imaging (MRI) and X-Ray scans. In this study, two separate datasets involving a total of 7600 MRI and X-Ray scans, available to users from Kaggle, were used. In the experimental part of this study, the 7600 MRI and X-Ray scans were trained and tested using a number of well-known learning models which are; XGBoost, Convolutional Neural Networks (CNN), ResNet50, DenseNet121 and AlexNet. After the experimental studies, the accuracy, sensitivity, specificity, precision, recall and F1-Score metrics of each of these models were obtained and compared. In addition, taking into account the results obtained from these models, a fusion approach named Majority Voting has been applied and the performance value of the system has been successfully increased. In summary, the performance results obtained from the used model are 98.98% for XGBoost, 99.75% for CNN, 99.65% for DenseNet121, 97.21% for ResNet50 and 99.84% for AlexNet. The accuracy after the Majority Voting approach applied is 100.00%. The results of the experimental studies demonstrate the promise of the proposed hybrid learning system with the Majority Voting approach and emphasizing its feasibility, effectiveness, and efficiency in processing both MRI and X-Ray tumor scanning techniques. Additionally, comparison with the state-of-the-art demonstrates that the proposed model outperforms the existing models for brain tumor detection.