Enhanced brain tumor classification using EfficientNetB0 and SVM with pareto search algorithm optimization

Abstract

Classifying tumors by type, grade, and stage is crucial for treatment decisions and predicting outcomes. Deep learning, especially Convolutional Neural Networks (CNNs), has significantly advanced tumor classification by effectively analyzing complex patterns in magnetic resonance (MR) images. This work presents a hybrid image classification method using the EfficientNetB0 model and Support Vector Machine (SVM) to categorize brain MR images into pituitary tumor, glioma tumor, meningioma tumor, and normal brain. EfficientNetB0 model extracts deep features from the images, which are then classified by a linear SVM. To significantly enhance classification accuracy for brain images, we use the Pareto algorithm to determine the penalty parameter C for the linear SVM. The testing results showed that the proposed system achieved a classification accuracy of 99.30%, recall of 99.30%, precision of 99.30%, and F1-score of 99.30%, with a high specificity of 99.77%. These results demonstrate that the combination of the Pareto algorithm and SVM significantly contributes to improved classification accuracy for brain images.