The healthcare sector has been overburdened with the massive outbreak of COVID-19 for almost two years and has created an urgent need for remote patient monitoring and treatment with minimal human involvement. Adopting artificial-intelligence- (AI)-based techniques for patient treatment has brought a dramatic revolution into the modern smart healthcare system (SHS). Modern healthcare is leveraging the internet of medical things (IoMT) network comprised of wireless body sensor devices (WBSDs) and implantable medical devices (IMDs) to reduce treatment latency and cost drastically. However, the open network communication of the less secured IoMT devices and increasing growth of adversarial capability give rise to several vulnerabilities, which need to be taken into consideration while designing SHS. We propose a novel and comprehensive framework with machine learning (ML) and formal analysis capability to build a secure and attack-resilient SHS. Our framework uses a novel ensemble of unsupervised ML-based patient status classification and anomaly detection models with bio-inspired computing (BIC)-based ML models’ hyperparameter optimization techniques. The proposed anomaly detection model (ADM) can detect zero-day attacks and uses a novel fitness function calculation technique for BIC-based SHS ADM’s hyperparameter optimization. Moreover, the framework leverages novel formal attack analytics to assess the robustness of the underlying classification and abnormality detection models. Our framework is evaluated using the University of Queensland Vital Signs dataset and a realistic synthetic dataset.
