A Comprehensive Review on Network Security in AI-based Healthcare Systems

Abstract

Artificial Intelligence (AI) has revolutionized healthcare with automated diagnosis and predictive analytics. However, the application of AI in healthcare systems introduces serious cybersecurity concerns, particularly concerning data integrity and model robustness. The robustness of Convolutional Neural Network (CNN), Support Vector Machine (SVM), and Random Forest models to adversarial attacks such as the Fast Gradient Sign Method (FGSM) and data poisoning is tested in this study. We take a hybrid experimental setting to estimate the performance of the model under clean, attacked and defense scenarios. The results show that CNN is highly vulnerable to adversarial attacks, while Random Forest is relatively more stable. While defence mechanisms such as adversarial training and knowledge distillation can improve the model’s performance, they cannot completely eliminate cybersecurity threats. The study emphasises the need for multi-layered cybersecurity frameworks. Such frameworks are necessary to guarantee the safety, privacy and reliability of AI-driven healthcare systems.