Explainable Deep Ensemble Meta-Learning Framework for Brain Tumor Classification Using MRI Images.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Machine Learning

Background:

• Brain tumors significantly impact neurological function and can be life-threatening if not detected early.
• Early detection of brain tumors is crucial to prevent permanent cognitive damage and improve patient outcomes.
• Magnetic Resonance Imaging (MRI) is a key modality for visualizing brain structures and detecting tumors.

Purpose of the Study:

• To develop an interpretable deep ensemble model for accurate brain tumor detection in MRI.
• To enhance classification accuracy and robustness using a stacking architecture with a meta-learner.
• To improve clinical trust through explainable AI methods for visualizing tumor regions.

Main Methods:

• Integration of pre-trained Convolutional Neural Networks (EfficientNetB7, InceptionV3, Xception) with a soft voting ensemble.
• Utilizing a Light Gradient Boosting Machine as a meta-learner within a stacking architecture.
• Employing hyperparameter tuning with Optuna and regularization techniques (batch normalization, L2 decay, dropout, early stopping, data augmentation) to prevent overfitting.

Main Results:

• The proposed framework achieved a high classification accuracy of 99.83% on a dataset of 3060 MRI images (BR35H dataset).
• Regularization strategies significantly improved the model's generalization ability.
• Explainable AI methods (Grad-CAM++, LIME, SHAP) successfully visualized tumor regions, enhancing interpretability.

Conclusions:

• The interpretable deep ensemble model demonstrates high accuracy and robustness in brain tumor detection from MRI scans.
• Explainable AI methods are vital for building clinical trust and understanding model predictions in medical imaging.
• This work provides a strong foundation for developing advanced AI-powered radiology decision support systems.