A Study on Enhancing Privacy in Medical Data Analysis through the Integration of Federated Learning's Privacy Limitations and Differential Privacy Techniques

연합학습의 개인정보 보호 한계와 차등 개인정보 보호 기법의 결합을 통한 의료 데이터분석 보안 향상 방안 연구

In this paper, Medical data contains sensitive personal information and health details about patients, making its secure protection a critical issue. Since medical data is used for purposes such as diagnosis, treatment, and research, it requires high accuracy and security. In the event of a data breach, there can be severe risks to patient privacy and health. Following the Fourth Industrial Revolution, medical data is increasingly analyzed through artificial intelligence, contributing significantly to the efficiency and accuracy of healthcare services. However, medical data requires stricter protective measures compared to general data, necessitating the adoption of new security technologies. This paper proposes a solution that combines Federated Learning and Differential Privacy to enable the secure analysis of medical data. Federated Learning reduces the risk of privacy breaches by sharing only the results of local data processing without centralizing the data on a server. However, it remains vulnerable to issues such as data imbalance and model inversion attacks. To address these limitations, Differential Privacy is applied by adding statistical noise to model updates, thereby reducing the risk of privacy infringement.