Automated Brain Tumor Classification Using RIE50: A Hybrid CNN and BLS Approach

Document Type : Original Article

Authors

1 Communication and Electronics Department Nile Higher Institute of Engineering and Technology, Mansoura 35511, Egypt

2 Electronics and Communications Engineering (ECE) Department, Faculty of Engineering, Mansoura University, Mansoura, 35516, Egypt

3 Electronics and Communication Engineering Department, Mansoura University, Mansoura 35516, Egypt

Abstract

Accurate and automated brain tumor classification is crucial for early detection and effective strategy development. This research introduces RIE50, a hybrid deep learning model that combines multiple pre-trained feature extractors with BLS-CNN layers to strengthen feature representation and boost classification accuracy. The BT-Large-4C dataset, used for model evaluation, comprises four categories: no tumor, glioma, pituitary tumor, and meningioma. The architecture incorporates dense layers of sizes 256, 512, and 1024, followed by batch normalization, dropout, and ReLU activation to enhance training stability and prevent overfitting. To optimize performance, we assessed two optimization techniques: SGDM and Adam, achieving classification accuracies of 97.12% and 97.54%, respectively. The experimental findings show that our approach effectively captures complex patterns in brain tumor images, leading to improved classification performance and enhanced model robustness. These results emphasize the promise of hybrid deep learning architectures in enhancing medical image analysis, aiding clinical decision-making, minimizing diagnostic errors, and optimizing patient outcomes

Keywords

Main Subjects

Files

Share

How to cite

Statistics