How can artificial intelligence be used in medical diagnostics while ensuring complete patient data privacy?
At the recent ETRAN Conference, Ana Vitković, a researcher at IMP-Automatika, presented the results of her research demonstrating that this is possible through the use of Fully Homomorphic Encryption (FHE).
In her paper, Privacy-Preserving Medical Diagnosis Using Homomorphic Encryption, she evaluated multiple combinations of machine learning models and FHE frameworks to identify the best balance between accuracy, computational efficiency and security when processing encrypted medical data.
Based on these findings, the HeartGuard AI web application was developed, enabling predictions to be performed on fully encrypted patient data. The application achieves 86.13% accuracy, produces predictions identical to those obtained from the same model running on unencrypted data, and performs server-side inference in just 15 milliseconds.
The research demonstrates that the future of medical diagnostics is not only about developing smarter AI models, but also about building technologies that enable advanced AI to operate without compromising patient privacy.
