An Explainable Artificial Intelligence Framework for Breast Cancer Detection

Breast cancer remains a leading cause of mortality among women worldwide, primarily due to delayed detection and a lack of early awareness. To address this issue, this study develops an advanced, thermal image-based breast cancer detection system that is non-invasive, radiation-free, and cost-effective, enhanced through the integration of artificial intelligence (AI) techniques. The proposed framework incorporates Attention U-Net for accurate segmentation of thermal breast images, K-Means Clustering to localize and isolate high-temperature regions suspected to be cancerous, and an EfficientNet-B7-based Convolutional Neural Network (CNN) for classification. To increase clinical reliability and transparency, the system employs Explainable AI (XAI) techniques using Local Interpretable Model-Agnostic Explanations (LIME), which provide visual interpretations of the model’s decision-making process. The dataset used in this research was obtained from the Database for Mastology Research (DMR) and consists of 2010 thermal images, including both healthy and abnormal cases. Preprocessing and segmentation effectively remove irrelevant areas and focus on the breast region, enhancing detection accuracy. Experimental evaluation indicates the proposed model achieves a training accuracy of 96.48% and a validation accuracy of 91.67%, with a recall of 91.95%, specificity of 91.43%, precision of 89.89%, and F1-score of 90.91%. These results highlight the system’s robust performance and generalizability. The LIME-generated superpixel visualizations help medical professionals better understand and validate the model's predictions, contributing to increased trust in AI-driven diagnostics. Overall, this research presents a reliable, explainable, and ethically grounded solution for early-stage breast cancer detection, demonstrating its strong potential for supporting clinical decision-making and future deployment in real-world healthcare settings.