AnomiGAN: Generative Adversarial Networks for Anonymizing Private Medical Data

Area of Science:

• Medical Informatics
• Machine Learning
• Genomic Data Security

Background:

• Personal medical data, including genomic sequences, contains sensitive information, posing risks during storage and sharing.
• Despite risks, medical data is crucial for advancing disease prevention and treatment research.
• Existing methods for data privacy may not adequately balance protection with predictive utility.

Purpose of the Study:

• To propose a novel framework, AnomiGAN, for preserving the privacy of personal medical data.
• To ensure high prediction performance is maintained alongside robust data privacy.
• To address the need for reliable techniques to prevent misuse of sensitive health information.

Main Methods:

• Development of AnomiGAN, a framework utilizing anonymized generative adversarial networks.
• Comparison of AnomiGAN against state-of-the-art privacy-preserving techniques.
• Validation of the model using datasets from the UCI machine learning repository.

Main Results:

• AnomiGAN achieves a privacy preservation level comparable to differential privacy (DP).
• The proposed method demonstrates superior prediction results compared to existing techniques.
• A trade-off between privacy levels and prediction performance was observed, dependent on data preservation.

Conclusions:

• AnomiGAN offers an effective solution for protecting sensitive medical data privacy.
• The framework successfully balances privacy preservation with high predictive performance.
• The study highlights the practical utility of AnomiGAN in real-world applications for medical data analysis.