Authors: Mrs. M. Sujana Priyadarshini, Vinnakoti Sakyavardhan
Abstract: Brain tumour diagnosis using Magnetic Resonance Imaging (MRI) plays a crucial role in early treatment planning and patient survival. However, manual interpretation of MRI scans is time-consuming and may lead to inconsistent clinical decisions. To address these limitations, this study proposes an explainable ensemble transfer learning framework for multiclass brain tumour classification. The proposed model integrates multiple pre-trained convolutional neural network architectures and aggregates their predictions using an ensemble strategy to enhance classification robustness and reduce overfitting. Furthermore, Explainable Artificial Intelligence (XAI) techniques are incorporated to visualize tumour regions and improve model interpretability, thereby increasing clinical trust and reliability. The dataset consists of multiclass MRI images categorized into glioma, meningioma, pituitary tumour, and no-tumour classes. Data augmentation and preprocessing techniques are employed to improve generalization performance. Experimental evaluation demonstrates that the ensemble framework achieves superior classification accuracy compared to individual transfer learning models. Performance is assessed using accuracy, precision, recall, F1-score, and confusion matrix analysis. The integration of explainability tools further validates the model’s capability to focus on clinically relevant tumour regions. The proposed approach offers a reliable, scalable, and interpretable solution for automated brain tumour detection and classification, making it suitable for real-world clinical decision support systems.
DOI: https://doi.org/10.5281/zenodo.19062217
