Developing a hybrid model for brain tumor detection using machine learning and deep learning techniques

Brain tumor detection from Magnetic Resonance Imaging (MRI) scans is a critical research area in medical imaging, as it directly affects patient outcomes, treatment planning, and early diagnosis processes. However, the manual evaluation of large and complex MRI data by radiologists is a time-consuming, subjective process with a high probability of error. In order to overcome these challenges, this thesis proposes a hybrid deep learning framework for brain tumor detection that integrates transfer learning–based pretrained Convolutional Neural Network (CNN) models, the Convolutional Block Attention Module (CBAM), and a Support Vector Machine (SVM) classifier. Five CNN architectures—ResNet50, InceptionV3, InceptionResNetV2, DenseNet121, and VGG16—are employed as feature extractors to derive fundamental patterns from two-dimensional brain MRI slices. To enhance the discriminative capability of the features, CBAM is integrated into the model, applying both spatial and channel-level attention mechanisms to enable the model to emphasize the most informative regions in brain tumor scans. The resulting enhanced features are transferred to the SVM classifier, which provides robust classification performance for brain tumor detection. The proposed method was developed using a dataset designed to distinguish between healthy and tumor data. The developed model was applied to two alternative datasets to evaluate its generalization performance. Additionally, three different data-splitting strategies were employed during the model validation phase. Experimental results demonstrate that the hybrid approach exhibits superior performance in brain tumor detection compared to traditional methods. In particular, the integration of the InceptionResNetV2 and DenseNet121 architectures with CBAM and SVM achieved the highest performance, attaining 99.77% accuracy and a 100% F1-score on the dataset. These findings confirm the potential of integrating attention mechanisms into clinical workflows and provide an effective and scalable solution for automated brain tumor classification. Keywords: Attention Mechanism, Brain Tumor, Convolutional Block Attention Module, Deep Learning, Hybrid Model, Transfer Learning, Support Vector Machine