Medical imaging is vital in the management of cancer especially during diagnosis and treatment selection as well as monitoring of patients during therapy. Radiologists primarily use magnetic resonance imaging to scan for brain tumors and automated classification remains challenging due to the shape, location and appearance of the tumor. Although CNN-based approaches have seen significant improvements, current methods lack interoperability and are often not optimal. To overcome this challenge this paper advances a novel approach using VGG19 with the integration of handcrafted feature fusion aimed at improving both classification accuracy and computational efficiency. Deep learning combined with handcrafted feature engineering fills an important gap in brain tumor classification. The objective is to design an automated system for the classification of abnormal components in brain magnetic resonance imaging, allowing radiologists to make correct and timely decisions that may improve patient outcome and reduce diagnostic delay. In an initial evaluation existing model such as VGG16, VGG19, ResNet50, and ResNet101, are compared with a SoftMax classifier. Moreover, a model is proposed that is based on deep feature-based classification, incorporating decision trees, SVM-RBF, and SVM-linear classifiers. The study further enhances the VGG19 model by incorporating handcrafted features, which results in improved accuracy. Experiments conducted using MRI slices show that the modified VGG19 achieves over 90% accuracy, along with reduced training time, thus highlighting its potential for real-world application in clinical settings.