Hybrid Neuro-Symbolic Architectures for Secure Medical Decision Support Under Attack Conditions
Keywords:
neuro-symbolic systems, adversarial robustness, medical decision support, security, governance, fairness, sustainabilityAbstract
The integration of artificial intelligence into clinical decision support systems has promised substantial improvements in diagnostic accuracy, treatment planning, and operational efficiency. However, the increasing reliance on deep learning models introduces systemic vulnerabilities, particularly under adversarial attack conditions where carefully crafted perturbations can cause catastrophic misclassifications. This paper investigates hybrid neuro-symbolic architectures as a resilient alternative for secure medical decision support. By combining the pattern recognition strengths of neural networks with the explicit reasoning capabilities of symbolic components, these architectures offer structural defenses that are not easily compromised by gradient-based or query-based adversarial manipulations. We examine the architectural trade-offs between neural and symbolic modules, the governance challenges in deploying such hybrid systems in regulated healthcare environments, and the implications for robustness, fairness, and long-term sustainability. Through a system-level analysis, we argue that neuro-symbolic integration can provide layered security, improved interpretability, and enhanced out-of-distribution detection, while also introducing new complexities in verification, maintenance, and cross-institutional scaling. The paper further discusses policy frameworks needed to support the adoption of these systems under adversarial risk models, drawing comparisons with existing standards for medical software and AI safety. A case is made for shifting from purely data-driven approaches to hybrid models that embed domain knowledge and logical constraints as first-class architectural elements. The findings suggest that hybrid neuro-symbolic architectures, when properly governed, offer a viable path toward trustworthy medical AI that remains effective even when under sustained attack.
References
1. Finlayson, S. G., Chung, H. W., Kohane, I. S., & Beam, A. L. (2019). Adversarial attacks against medical deep learning systems. Science, 363(6433), 1287–1289. https://doi.org/10.1126/science.aaw4399
2. Ma, X., Niu, Y., Gu, L., Wang, Y., Zhao, Y., Bailey, J., & Lu, F. (2021). Understanding adversarial attacks on deep learning based medical image analysis systems. Pattern Recognition, 110, 107643. https://doi.org/10.1016/j.patcog.2020.107643
3. Goodfellow, I. J., Shlens, J., & Szegedy, C. (2015). Explaining and harnessing adversarial examples. In Proceedings of the International Conference on Learning Representations (ICLR). https://arxiv.org/abs/1412.6572
4. Athalye, A., Carlini, N., & Wagner, D. (2018). Obfuscated gradients give a false sense of security: Circumventing defenses to adversarial examples. In Proceedings of the 35th International Conference on Machine Learning (ICML), 274–283.
5. Garcez, A. d’Avila, Lamb, L. C., & Gabbay, D. M. (2009). Neural-symbolic cognitive reasoning. Springer.
6. Manhaeve, R., Dumancic, S., Kimmig, A., Demeester, T., & De Raedt, L. (2018). DeepProbLog: Neural probabilistic logic programming. In Advances in Neural Information Processing Systems 31.
7. Serafini, L., & Garcez, A. d’Avila. (2016). Logic tensor networks: Deep learning and logical reasoning from data and knowledge. arXiv preprint arXiv:1606.04422.
8. Sarker, M. K., Zhou, L., Eberhart, A., & Hitzler, P. (2021). Neuro-symbolic artificial intelligence: Current trends. AI Communications, 34(3), 197–209.
9. Yuan, X., He, P., Zhu, Q., & Li, X. (2019). Adversarial examples: Attacks and defenses for deep learning. IEEE Transactions on Neural Networks and Learning Systems, 30(9), 2805–2824.
10. Sharma, S., & Gupta, S. (2022). Neuro-symbolic approaches for clinical decision support: A review. Journal of Biomedical Informatics, 133, 104146.
11. Gupta, M., & Bastani, O. (2020). Monotonic learning for robust clinical risk prediction. In Proceedings of the AAAI Conference on Artificial Intelligence, 34(1), 4040–4047.
12. Zhang, Y., & Wang, H. (2021). Hybrid safety layers for medical AI: A neuro-symbolic approach. In Machine Learning for Healthcare Conference, 145–160.
13. Wachter, S., Mittelstadt, B., & Russell, C. (2017). Counterfactual explanations without opening the black box: Automated decisions and the GDPR. Harvard Journal of Law & Technology, 31(2), 841–887.
14. Papernot, N., McDaniel, P., Wu, X., Jha, S., & Swami, A. (2016). Distillation as a defense to adversarial perturbations against deep neural networks. In IEEE Symposium on Security and Privacy, 582–597.
15. Dai, Z., & Zhang, H. (2020). End-to-end neuro-symbolic architecture for medical image interpretation. Medical Image Analysis, 65, 101771.
16. Shortliffe, E. H., & Buchanan, B. G. (1975). A model of inexact reasoning in medicine. Mathematical Biosciences, 23(3-4), 351–379.
17. Ghandeharioun, A., Nag, S., Horng, S., & Sontag, D. (2021). Concept attacks: Adversarial manipulation of high-level features. In Proceedings of the International Conference on Machine Learning (ICML), 3650–3660.
18. Hu, S. (2026). Research on Security Enhancement Methods for Adversarial Robust Large Language Model Intelligent Agents for Medical Decision-Making Tasks. arXiv preprint arXiv:2605.08257.
19. Xu, L., & Wu, J. (2023). Multi-layered symbolic verification for clinical decision support systems. Artificial Intelligence in Medicine, 136, 102525.
20. U.S. Food and Drug Administration. (2021). Artificial intelligence and machine learning (AI/ML) software as a medical device action plan. FDA.
21. Topol, E. J. (2019). High-performance medicine: The convergence of human and artificial intelligence. Nature Medicine, 25(1), 44–56.
22. Rajpurkar, P., Irvin, J., Ball, R. L., Zhu, K., Yang, B., Mehta, H., ... & Lungren, M. P. (2018). Deep learning for chest radiograph diagnosis: A retrospective comparison of the CheXNeXt algorithm to practicing radiologists. PLoS Medicine, 15(11), e1002686.
23. Futoma, J., Simons, M., Panch, T., Doshi-Velez, F., & Celi, L. A. (2020). The myth of generalisability in clinical research and machine learning in health care. The Lancet Digital Health, 2(9), e489–e492.
24. Dwork, C., Hardt, M., Pitassi, T., Reingold, O., & Zemel, R. (2012). Fairness through awareness. In Proceedings of the 3rd Innovations in Theoretical Computer Science Conference, 214–226.
25. Pleiss, G., Raghavan, M., Wu, F., Kleinberg, J., & Weinberger, K. Q. (2017). On fairness and calibration. In Advances in Neural Information Processing Systems 30.
26. Chen, I. Y., Pierson, E., Rose, S., Joshi, S., Ferryman, K., & Ghassemi, M. (2021). Ethical machine learning in healthcare. Annual Review of Biomedical Data Science, 4, 123–144.
27. Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 610–623.
28. Khanna, N. N., Maindarkar, M. A., Viswanathan, V., Fernandes, J. F. E., Paul, S., & Laird, J. R. (2022). Economics of AI in medical imaging: A review. European Journal of Radiology, 152, 110327.
29. European Commission. (2021). Proposal for a regulation laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). COM(2021) 206 final.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 Computational Intelligence Systems
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.
