Enhancing Privacy in Collaborative Breast Cancer Diagnosis: A Federated Learning Approach with Homomorphic Encryption
Vankamamidi Srinivasa Naresh, Gadhiraju Tej Varma, D Ayyappa · 2025
Cancer remains as one of the most prevalent and life threatening disease worldwide, triggering the need for advanced early diagnostic methods. At the same time protecting patients privacy is of greatest importance in healthcare as the critical medical data is being integrated with artificial intelligence diagnostic tools. In breast cancer detection, safeguarding personal health information is crucial to ensure adherence to legal and ethical standards while fostering trust and compliance. An innovative approach for preservation of privacy in federative learning while eliminated the need of maintaining the centralized data. The proposed model improves the privacy preservation of federative learning by implementing Homomorphic Encryption, which encrypts the model update during transmission to prevent unauthorized access of sensitive information. The proposed framework for breast cancer detection incorporates Federative Learning across five nodes, each node containing a dataset of at least 1000 samples with 32 features each. A Logistic regression model is trained collaboratively without sharing the data to a centralized server, ensuring confidentiality and data security. To further improve the model privacy, CKKS homomorphic encryption scheme is implemented through the TenSEAL library. This ensure utmost privacy as computation is also performed on the encrypted data this safeguarding the sensitive medical information. The proposed approach achieves an accuracy ranging from 78% to 83% in breast cancer detection, demonstrating the effectiveness of the secure and collaborative learning approaches.