An expert voting system for brain tumor Classification using MRI images

In the recent era of deep networks, the success of medical imaging has almost no limits, but the system proposed in this manuscript goes beyond imitating the expertise and decision-making of multiple radiologists. The study presents a voting system designed to enhance the accuracy of brain tumor Classification through the analysis of MRI images by separate classifiers. The system uses various classifiers trained on different data representations: Support Vector Machines (SVM) and K-Nearest Neighbors (KNN) using Histogram of Oriented Gradients (HOG) features, ResNet50 using edge images (see Canny edge detector), and Convolutional Neural Networks (CNN) using the original MRI images. Each of these classifier brings unique view/decision to the ensemble, capturing different aspects of the brain scans. The voting system aggregates the predictions from these diverse models through a weighted voting mechanism, where weights are assigned based on the individual performance of each classifier. This solution balances the strengths and diminish the weaknesses of each model, leading to improved generalization and robustness. Our experimental results demonstrate that the voting system significantly outperforms the individual classifiers and other ensemble methods, achieving a remarkable accuracy rate of 98.36%. The proposed solution proves the efficacy of ensemble methods in medical image Classification and demonstrates the potential of the proposed system to deliver more reliable and precise diagnostic assistance in clinical environments, thereby improving treatment strategies and patient care.