A hybrid CNN-SVM model optimized with PSO for accurate and non-invasive brain tumor classification

Brain and central nervous system (CNS) cancers are the leading cause of cancer-related mortality, presenting significant diagnostic challenges due to their aggressive nature and diverse manifestations. While biopsies are the gold standard for diagnosis, their invasiveness and associated risks necessitate the development of noninvasive, automated tools for accurate and rapid tumor classification. This research presents a hybrid approach that combines a convolutional neural network (CNN) for feature extraction with a support vector machine (SVM) classifier, optimized by particle swarm optimization (PSO). The proposed model achieves multiclass classification across four tumor categories: glioma, meningioma, pituitary, and no tumor. The model incorporates data augmentation, dropout layers, pooling, batch-normalization, and hyperparameter tuning via PSO, to enhance generalization and minimize overfitting. The experimental analysis on the brain tumor MRI dataset highlights the model's strong performance, with an overall accuracy of 84.77%, a macro-average F1-score of 0.80, and a macro-average precision of 0.91. The performance of the PSO-optimized CNN-SVM model was benchmarked against traditional machine learning (ML) and deep learning (DL) methods, which demonstrated mean accuracies between 48 and 72%. The proposed hybrid model significantly surpasses these conventional approaches, establishing its superiority in classification accuracy and reliability. The results highlight the proposed model's reliability as a noninvasive diagnostic solution, offering considerable upside for enhancing early brain tumor detection and aiding in treatment planning.