Hybrid Approach for the Classification and Detection of Brain Tumor
Year: 2026
Type: dissertation
Abstract: Brain tumors represent one of the most severeneurological disorders and require early detection for effectivetreatment. MRI imaging is widely used because it providesdetailed visualization of soft tissues and tumor structures withoutradiation exposure. Despite its advantages, interpreting MRIscans manually is complex and prone to inter-observer variability.Recent advancements in Artificial Intelligence (AI) and DeepLearning have enabled automated medical image analysis withhigh accuracy. Convolutional Neural Networks (CNNs) havedemonstrated strong performance in feature extraction and classification tasks. However, standalone CNN models often requirelarge datasets and may suffer from overfitting when applied tolimited medical datasets. In recent years, Artificial Intelligence(AI) and Deep Learning have significantly transformed medicalimage analysis. Convolutional Neural Networks (CNNs) havedemonstrated remarkable success in various computer visiontasks, including object detection, segmentation, and classification.In medical imaging, CNN-based architectures have been appliedto detect tumors, segment lesions, and classify disease typeswith high accuracy. Despite these advancements, standalone deeplearning models often face challenges such as overfitting, highcomputational complexity, and limited interpretability. Hybridmodels that combine deep learning feature extraction with classical machine learning classifiers have shown promising resultsin improving generalization performance. Additionally, segmentation of tumor regions before classification can significantlyimprove prediction accuracy by eliminating irrelevant background information. This work proposes a comprehensive hybridarchitecture that integrates: • U-Net for tumor segmentation• MobileNetV2 for transfer learning-based feature extraction• SVM and Random Forest for classification • Grad-CAM forexplainability The proposed approach aims to balance accuracy,computational efficiency, and interpretability in a unified framework.
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7154615453