Hybrid machine learning model towards neuroimaging analysis for detection and grading brain tumors

Abstract

Neuroimaging analysis enables detailed observation of brain tumors, with growing adoption of advanced imaging techniques in clinical practice. Limitations of conventional approaches in supporting proactive decisionmaking and reliable grading are increasingly addressed through machine learning. However, earlier models often faced challenges of limited generalization and computational burden. To overcome these issues, this study introduces a hybrid convolution neural network–support vector machine (CNN–SVM) framework that combines ResNet-50 feature extraction with a feature weighting (FW) strategy and SVM-based classification for improved diagnostic precision. The system is further enhanced with a clinically guided grading scheme, mapping classification outputs into malignant, benign, and healthy categories for greater interpretability. The proposed model was evaluated on three benchmark neuroimaging datasets (Figshare, Kaggle magnetic resonance imaging (MRI), and BraTS-2019) and achieved up to 98.6% accuracy with high sensitivity and specificity, while retaining low computational cost and rapid inference, outperforming conventional CNN-only methods.