Lung Cancer Detection with Privacy Preservation Using Federated Learning

Privacy protection in health research is a critical challenge, especially with centralized data storage solutions that expose sensitive information. This paper addresses this issue by proposing a Federated Learning (FL) model integrated with Secure Aggregation (SecAgg) for lungs cancer detection by the CT scan image. Unlike traditional FL methods, which rely on a central serv-er for model aggregation and risk data reconstruction, the proposed architecture employs Secure Multi-Party Computation (SMPC) techniques to decentralize aggregation. This ensures that raw client data remains private, significantly reducing privacy threats from malicious or semi-honest aggregators. The system uses a dataset of 1,097 CT scan images, classifying cases into malignant, benign, or normal with an achieved accuracy of 96.40 %. Evaluation metrics such as precision, recall, and F1-score further demonstrate the robustness of the proposed system, with precision reaching <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1. 0 0}$</tex> for benign cases and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 9 9}$</tex> for malignant cases. By eliminating dependency on a central server, this approach enhances data security while maintaining high model performance. The architecture represents a significant advancement in privacy-preserving federated learning for medical applications, with potential for broader deployment. Future work aims to enhance scalability and address remaining security challenges for wider adoption.