Explainable CNN for brain tumor detection and classification through XAI based key features identification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Existing brain tumor classification models often exhibit complex structures, hindering interpretability and potentially leading to reliance on irrelevant features.
• Model complexity increases the number of layers and parameters, complicating the classification process and reducing transparency.

Purpose of the Study:

• To develop a novel methodology combining Explainable AI (XAI) techniques with a simplified Convolutional Neural Network (CNN) architecture for brain tumor classification.
• To enhance model transparency and robustness by ensuring focus on relevant features and reducing complexity.

Main Methods:

• Integration of XAI techniques, including Gradient-weighted Class Activation Mapping (Grad-Cam), Shapley Additive explanations (Shap), and Local Interpretable Model-agnostic Explanations (LIME), with a CNN architecture.
• Minimization of network layers to reduce model complexity and improve focus on critical features for tumor detection.

Main Results:

• The proposed model achieved 99% accuracy on seen data and 95% accuracy on unseen data, demonstrating strong generalizability and reliability.
• XAI techniques provided clear insights into the model's decision-making process, confirming focus on relevant tumor features.

Conclusions:

• The developed approach offers a balance of simplicity, interpretability, and high accuracy in brain tumor classification.
• This methodology represents a significant advancement, improving transparency and reliability in medical image analysis for brain tumor detection.