A HYBRID DEEP LEARNING MODEL TO AUTOMATE THE CLASSIFICATION AND SEGMENTATION OF GLIOMA TUMORS FROM MRI MEDICAL IMAGES

Gliomas are among the most aggressive brain tumors experienced and the accurate classification of such disease is critical for effective diagnosis and treatment in patients. Therefore, this study proposes a hybrid deep learning model integrating Convolutional Neural Networks (CNNs), ResNet50, and U-Net to automate the classification and segmentation of glioma tumors from Medical Resonance Imaging (MRI) scans. The proposed model applies CNN for feature extraction in the pipeline, ResNet50 for deep residual learning and U-Net for precise pixel-level tumor localization and a dataset of 10,694 MRI images from 184 subjects, including both primary hospital data and secondary sources from the Roboflow repository, was used for training and validation. The dataset covered four glioma types: Glioblastoma Multiforme (GBM), Meningioma, Ependymomas, and Mixed Glioma. The model was trained using the Adam optimizer with categorical cross-entropy loss over 50 epochs, applying data augmentation to enhance robustness.Experimental results demonstrate that the hybrid model achieved a training accuracy of 95% and validation accuracy of 94%, outperforming standalone CNN and ResNet50 models. Confusion matrix analysis confirmed reliable multi-class classification, while practical testing validated accurate tumor segmentation and labelling. These results indicate that the proposed model provides a robust, scalable, and clinically applicable tool for automated glioma detection, supporting radiologists in diagnostic decision-making.