Breast Cancer Detection with MobileNetV2 and Explainable AI

To save lives, early identification of breast cancer is essential. While medical imaging benefits greatly from the use of models such as AlexNet, VGGNet and ResNet their high computing demands make their adoption difficult in locations with limited resources. This study presents an innovative approach to breast cancer detection by leveraging MobileNetV2, a model designed for efficiency in low-resource environments. While the standard MobileNetV2 achieved an accuracy level between 85% and 87%, our implementation of a custom classification head significantly enhanced performance, achieving an impressive accuracy of 98.8%. The model also demonstrated a precision of 97.5%, recall of 98.2%, and an Fl score of 97.8%. To further improve transparency, we applied Local Interpretable Model-agnostic Explanations (LIME), which gave us significant understanding into the model's decision-making process. This integration of efficiency and interpretability offers a viable answer to affordable breast cancer detection, with substantial potential for application in resource-limited settings.