Hybrid Deep–Classical Models for Brain Tumor Classification and Diagnosis

New approaches for diagnosing complex diseases with aid of computers such as artificial intelligence and CT scans have resulted from the recent developments in medical imaging as well as artificial intelligence. Deep learning architectures are one of the key strengths in feature learning; however, classical machine learning algorithms provide interpretability and computational efficiency besides their lesser accuracy. A model called Hybrid Deep-Classical Model is elaborated on, which consists of deep feature extraction by utilizing CNN architectures (VGG16, ResNet50) combined with classical classifiers like Support Vector Machine (SVM) and Random Forest (RF). The combination leads to an increase in accuracy and generalization particularly in the case of small medical datasets. The experiments conducted on the BRATS 2020 and Kaggle Brain MRI datasets show that the results improve with the hybrid model having an average accuracy of 97,8 %, precision of 96,9 %, and F1-score of 97,2 % respectively. It can thus be concluded from the results that the hybrid models are superior to the others in the case of biomedical imaging for the purpose of obtaining reliable and efficient diagnosis of diseases.