XAI-Driven Mammogram Classification: A Comparative Study of SHAP and LIME in Deep Learning Models

Breast cancer remains one of the leading causes of mortality among women worldwide, where early and accurate diagnosis plays a crucial role in improving survival rates. Mammography is the most widely used screening technique; however, its interpretation is often challenging due to subtle variations in tissue structures and high inter-observer variability among radiologists. In recent years, deep learning models, particularly Convolutional Neural Networks (CNNs), have demonstrated strong performance in automated mammogram classification. Despite their effectiveness, these models lack transparency, limiting their adoption in clinical decision-making due to their black-box nature. This study proposes an Explainable Artificial Intelligence (XAI)-driven framework for mammogram classification, focusing on a comparative analysis of SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-Agnostic Explanations). A ResNet-based CNN model is trained on the CBIS-DDSM dataset to classify mammograms into benign and malignant categories. SHAP and LIME are integrated to provide both global and local interpretability of model predictions. Experimental results indicate that while both methods enhance model transparency, SHAP provides more stable and consistent feature attributions across samples, whereas LIME offers faster and more localized explanations. Quantitative evaluation using fidelity and stability metrics, along with qualitative visualization analysis, highlights key differences between the two approaches. The findings suggest that SHAP is more suitable for clinical validation and auditing, while LIME is effective for rapid case-based interpretation. This work contributes toward improving trust and reliability in AI-assisted breast cancer diagnosis.