Privacy Preserving Breast Cancer Prediction with Mammography Images Using Federated Learning

Breast cancer is a global health concern demanding early detection for improved outcomes. Mammography identifies breast abnormalities, including early-stage cancer. Privacy concerns hinder predictive models using mammographic data. This thesis employs federated learning to predict breast cancer risk while preserving patient privacy. It allows collaborative model training across healthcare organizations without revealing raw data. Using diverse data, accurate risk prediction models consider parameters like breast density, microcalcifications, masses, architectural distortion, asymmetry, and prior mammograms. Clinical and demographic data enables tailored screening. The study leverages federated learning and mammographic components for precise, privacy-conscious risk prediction. Addressing privacy issues in mammography data enables personalized care. Integrating genetic, clinical, and mammographic data may lead to precision medicine. The goal is to improve patient quality of life, reduce mortality, and enhance early detection. With a dataset of four classes and 6,649 images, the model achieves 72.46% accuracy, laying the foundation for advanced privacy-preserving risk prediction models and personalized care.