Privacy-Preserving Deep Neural Inference for Clinical Prediction Using Adaptive Polynomial Activation Approximation and Encrypted Medical Data
Keywords:
homomorphic encryption; clinical prediction; polynomial activation approximation; encrypted neural inference; noise-aware planning; decision calibrationAbstract
Clinical prediction models increasingly rely on deep neural networks trained on electronic health records, laboratory values, vital signs, and imaging-derived features. Although such models can capture nonlinear interactions relevant to risk stratification, cloud-hosted or third-party inference exposes sensitive medical data and intermediate representations. Homomorphic encryption offers a principled way to evaluate a network on encrypted inputs, but its algebraic limitations make nonlinear activations difficult to compute. Fixed polynomial replacements can work on a nominal interval, yet they may fail when activation distributions shift across patient subgroups, clinical sites, or layers. This manuscript develops a privacy-preserving inference framework that couples encrypted medical data with adaptive polynomial activation approximation. The method selects layer-specific approximation intervals, polynomial degrees, and encrypted normalization parameters by minimizing distribution-weighted approximation error under a noise-depth budget. It then propagates approximation error through the network to obtain a risk-score perturbation bound and links that bound to clinical decision margins and abstention rules. The analysis clarifies how adaptive degree allocation differs from uniform polynomial replacement, why activation-range normalization is important for deep encrypted inference, and how leakage boundaries arise from ciphertext access patterns, approximation statistics, and model architecture. The contribution is methodological and analytical: it provides a coherent design for encrypted clinical prediction, derives explicit trade-offs among activation fidelity, multiplicative depth, and decision stability, and identifies failure modes that must be controlled before such models can be used in privacy-sensitive clinical workflows.
References
1. Gilad-Bachrach, R., Dowlin, N., Laine, K., Lauter, K., Naehrig, M., & Wernsing, J. (2016). CryptoNets: Applying neural networks to encrypted data with high throughput and accuracy. In Proceedings of the 33rd International Conference on Machine Learning (PMLR 48, pp. 201–210).
2. Cheon, J. H., Kim, A., Kim, M., & Song, Y. (2017). Homomorphic encryption for arithmetic of approximate numbers. In Advances in Cryptology – ASIACRYPT 2017 (LNCS 10624, pp. 409–437). Springer.
3. Bourse, F., Minelli, M., Minihold, M., & Paillier, P. (2018). Fast homomorphic evaluation of deep discretized neural networks. In Advances in Cryptology – CRYPTO 2018 (LNCS 10993, pp. 483–512). Springer.
4. Chabanne, H., de Wargny, A., Milgram, J., Morel, C., & Prouff, E. (2017). Privacy-preserving classification on deep neural network. Cryptology ePrint Archive, Report 2017/035.
5. Hesamifard, E., Takabi, H., & Ghasemi, M. (2017). CryptoDL: Deep neural networks over encrypted data. arXiv:1711.05189.
6. Brutzkus, A., Gilad-Bachrach, R., & Erez, S. (2019). Low latency privacy preserving inference. In Proceedings of the 36th International Conference on Machine Learning (PMLR 97, pp. 812–821).
7. Miotto, R., Li, L., Kidd, B. A., & Dudley, J. T. (2016). Deep patient: An unsupervised representation to predict the future of patients from the electronic health records. Scientific Reports, 6, 26094.
8. Rajkomar, A., Oren, E., Chen, K., Dai, A. M., Hajaj, N., Hardt, M., Liu, P. J., Liu, X., Marcus, J., Sun, M., Sundberg, P., Yee, H., Zhang, K., Duggan, G. E., Flores, G., Laramie, D., Li, Y., Long, X., Lu, J., … Dean, J. (2018). Scalable and accurate deep learning with electronic health records. NPJ Digital Medicine, 1, 18.
9. Shickel, B., Tighe, P. J., Bihorac, A., & Rashidi, P. (2018). Deep EHR: A survey of recent advances in deep learning techniques for electronic health record (EHR) analysis. IEEE Journal of Biomedical and Health Informatics, 22(5), 1589–1604.
10. Fredrikson, M., Jha, S., & Ristenpart, T. (2015). Model inversion attacks that exploit confidence information and basic countermeasures. In Proceedings of the 22nd ACM SIGSAC Conference on Computer and Communications Security (pp. 1322–1333). ACM.
11. Shokri, R., Stronati, M., Song, C., & Shmatikov, V. (2017). Membership inference attacks against machine learning models. In 2017 IEEE Symposium on Security and Privacy (pp. 3–18). IEEE.
12. Nasr, M., Shokri, R., & Houmansadr, A. (2019). Comprehensive privacy analysis of deep learning: Passive and active white-box inference attacks against centralized and federated learning. In 2019 IEEE Symposium on Security and Privacy (pp. 739–753). IEEE.
13. Papernot, N., Abadi, M., Erlingsson, U., Goodfellow, I., & Talwar, K. (2017). Semi-supervised knowledge transfer for deep learning from private training data. In International Conference on Learning Representations.
14. Chillotti, I., Gama, N., Georgieva, M., & Izabachène, M. (2020). TFHE: Fast fully homomorphic encryption over the torus. Journal of Cryptology, 33(1), 34–91.
15. Paterson, M. S., & Stockmeyer, L. J. (1973). On the number of nonscalar multiplications necessary to evaluate polynomials. SIAM Journal on Computing, 2(1), 60–66.
16. Lou, Q., & Jiang, L. (2019). SHE: A fast and accurate deep neural network for encrypted data. In Advances in Neural Information Processing Systems 32.
17. Kim, M., Song, Y., Wang, S., Xia, Y., & Jiang, X. (2018). Secure logistic regression based on homomorphic encryption: Design and evaluation. JMIR Medical Informatics, 6(2), e19.
18. Peng, H., Huang, S., Zhou, T., Luo, Y., Wang, C., Wang, Z., ... & Ding, C. (2023, October). Autorep: Automatic relu replacement for fast private network inference. In 2023 IEEE/CVF International Conference on Computer Vision (ICCV) (pp. 5155-5165). IEEE.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 International Journal of Clinical and Translational Medicine

This work is licensed under a Creative Commons Attribution 4.0 International License.
This article is published under the Creative Commons Attribution 4.0 International License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.



