Enhancing Deep Learning Model Explainability in Brain Tumor Datasets Using Post-Heuristic Approaches.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Machine Learning

Background:

• Deep learning models show high efficacy in medical diagnosis.
• A key limitation is the lack of explainability in their decision-making processes.
• Enhancing model interpretability is crucial for clinical adoption.

Purpose of the Study:

• To improve the robustness and interpretability of deep learning models in medical diagnosis.
• To refine explanations generated by the LIME (Local Interpretable Model-agnostic Explanations) Library and LIME image explainer.
• To develop a post-heuristic approach for more concrete diagnostic explanations.

Main Methods:

• Utilized post-processing mechanisms based on scenario-specific rules.
• Applied these methods to refine explanations from the LIME Library and LIME image explainer.
• Conducted experiments using publicly accessible brain tumor detection datasets.

Main Results:

• The proposed post-heuristic approach significantly advanced the interpretability of deep learning models.
• Achieved more robust and concrete explanations in the context of medical diagnosis.
• Demonstrated improved reliability for AI-driven diagnostic tools.

Conclusions:

• The developed method enhances the trustworthiness of deep learning in medical diagnosis.
• Scenario-specific rule-based post-processing offers a viable solution to the explainability problem.
• This research contributes to more reliable and interpretable AI applications in healthcare.