Ensemble-based Convolutional Neural Networks for brain tumor classification in MRI: Enhancing accuracy and...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Accurate brain tumor classification (gliomas, meningiomas, pituitary adenomas) is vital for patient diagnosis and treatment.
• Magnetic resonance imaging (MRI) is a primary diagnostic modality.
• Deep learning shows potential for automated tumor classification but faces challenges in accuracy and clinical interpretability.

Purpose of the Study:

• To enhance the accuracy and interpretability of brain tumor classification using deep learning on MRI images.
• To develop a robust and clinically applicable AI tool for supporting medical professionals.

Main Methods:

• Utilized transfer learning with pre-trained Convolutional Neural Network (CNN) architectures (VGG16, DenseNet121, Inception-ResNet-v2).
• Developed an ensemble classifier with a majority voting strategy for improved robustness.
• Integrated explainability techniques (Grad-CAM++, Integrated Gradients) for visualizing model decisions.

Main Results:

• The ensemble model achieved 86.17% accuracy in classifying gliomas, meningiomas, pituitary adenomas, and benign cases.
• Explainability methods generated heatmaps highlighting critical regions for predictions, aligning with radiological features.
• The ensemble approach demonstrated superior performance compared to individual CNN architectures.

Conclusions:

• The proposed ensemble deep learning framework significantly improves brain tumor classification accuracy and interpretability from MRI.
• Integrating explainability methods enhances the transparency and reliability of AI diagnostic tools.
• This approach provides valuable support for medical professionals in clinical decision-making.