Explainable deep learning framework for brain tumor detection: Integrating LIME, Grad-CAM, and SHAP for enhanced...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• Deep learning enhances disease diagnosis but often lacks transparency.
• AI decision systems require improved interpretability for clinical trust.
• Brain tumor detection is a critical application for AI in healthcare.

Purpose of the Study:

• To enhance the explainability and training performance of deep learning models for brain tumor detection.
• To integrate explainable artificial intelligence (XAI) techniques into a convolutional neural network (CNN).
• To validate the effectiveness of XAI in improving diagnostic accuracy and model reliability.

Main Methods:

• A two-stage training approach was employed for a CNN model.
• Explainable AI (XAI) methods including Local Interpretable Model-Agnostic Explanations (LIME), Gradient-weighted Class Activation Mapping (Grad-CAM), and Shapley Additive Explanations (SHAP) were utilized.
• Generated XAI masks were used to augment the dataset for improved training.

Main Results:

• The initial CNN model achieved 97.20% accuracy on the BRATS2019 dataset.
• Integration of LIME, Grad-CAM, and SHAP masks improved accuracy to 99.40%, with sensitivity at 99.20%, specificity at 99.60%, and ROC-AUC at 99.90%.
• The model demonstrated generalizability, with accuracy increasing from 96.80% to 99.80% on the BR35H dataset.

Conclusions:

• XAI techniques significantly enhance the performance and interpretability of deep learning models for brain tumor detection.
• The proposed method offers a reliable and stable strategy for improving AI diagnostic accuracy in medical imaging.
• The integration of XAI is crucial for developing trustworthy and effective AI-driven clinical decision support systems.