Explainability in CNN based Deep Learning models for medical image classification

DL techniques have increased the efficiency of decision making in different areas. However, in the case of the presence of uncertainties in the data or in the environment, decision-making requires the explainability of the model, especially for high-stakes decision making such as medical image analysis area. This paper focuses on medical image classification CNN based deep learning models and aims to apply and compare three popular explainable AI approaches LIME, SHAP and GradCAM.Simulation results on a Pneumonia and Alzheimer’s datasets for disease detection show that the Grad-CAM method seems to outperform LIME and SHAP and able to enhance the interpretability of DL models, identify automatically the most important features that contribute to the model’s decision.