DEVELOPMENT OF A REAL-TIME VULNERABILITY MANAGEMENT SYSTEM FOR HEALTHCARE CYBER-PHYSICAL SYSTEMS USING A HYBRID DEEP LEARNING

The growth of the utilization of Internet of Things (IoT) and cyber-physical systems (CPS) in the healthcare sector has come along with new vulnerabilities in cybersecurity, which endanger the safety and integrity of patients and data. Considering that a hybrid deep learning framework based on Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) networks is developed in the context of this study, this paper introduces real-time vulnerability management system of medical cyber-physical systems on the hybrid deep learning framework. The system was tested and trained based on the data of vulnerabilities taken at the University of Nigeria Teaching Hospital (UNTH), one of the targeted intensive care units (ICU) devices between 2019 and 2022. The data was also extensively pre-processed (number of missing values was imputed, the normalization and class balancing with random under-sampling was performed). The CNN layer allowed extracting spatial features whereas the LSTM layer allowed capturing temporal patterns in network traffic and system logs. It was written in Python, TensorFlow, and Keras with tools that allow real-time scanning included, namely ClamAV and Nmap. Evaluation of the performance has shown high precision (97 training and 93 validation), and the F1-score was 0.94 due to the good result of the system in identifying and categorizing the known and upcoming vulnerabilities. The system was found to be real-time deployable as well as providing proactive threat detection at levels of multiple networks. The study would help enhance cybersecurity resilience in healthcare, as the method relied on by this research allowed early warning against and remediation of risks within mission-critical medical settings.