Brain Tumor Classification Using Enhanced CNN and Optimization with Metaheuristic Algorithms

Brain tumors are now the second most common serious consequence worldwide, after cardiovascular disorders. Early detection and treatment of brain cancers save millions of lives worldwide. Current ways to evaluate brain tumors need human intervention, which is time-consuming and subjective. Automated brain tumor detection methods are critical for a speedy and accurate diagnosis. Existing approaches are frequently intrusive, emphasizing the necessity for a fully automated deep-learning system to classify brain cancers. In this study, we employed Convolutional Neural Networks (CNNs) and five pre-trained models—ResNet-101, VGG-19, MobileNet, XceptionNet, and Inception V3—on 7,800 MRI images and evaluated each across several epochs. In comparison, it is found that these models achieve a greater accuracy at 50 epochs. The models experienced overfitting as they progressed to higher epochs, resulting in saturated accuracy. To further increase the accuracy, the weights of the pretrained models were optimized using the metaheuristic algorithms such as PSO, GA and ACO. The optimized Inception V3 model outperformed other pre-trained models with F1 scores of approximately 98.2%. These results demonstrate the efficacy of our approach in providing rapid and accurate brain tumor classification, addressing the pressing need for reliable diagnostic tools in clinical settings.