Keywords

1 Introduction

1.1 Overview

The recent development in cutting edge technologies like Internet of Things (IoTs), Digital Twins (DTs), augmented reality, and advanced data analytics present invaluable applications in many critical domains such as air traffic monitoring, e-health, smart cities, and smart grids, among others. For these critical domains, Information Technology (IT) and Operational Technology (OT) are significant for optimization and customization of their respective functions. The convergence of IT and OT introduces specific cyber security challenges, like the combination of legacy OT systems with modern IT networks that expose difficult-to-patch vulnerabilities, lack standardization, increase attack surface due to integration complexity, thereby creating potential pathways for cyber threats to move between the two types of technology. The security challenges raised by IT-OT integration require real-time and secured communication between different stakeholders. Meanwhile, attackers are using advanced technology, such as Artificial Intelligence (AI)/Machine Learning (ML) and behavioral biometrics, to disrupt the regular operations of healthcare sectors. Therefore, there is a need to develop an approach to effectively monitor, predict, respond to and recover from security threats against different healthcare applications in a timely manner that confines any negative consequences [1].

To achieve a high-level of cyber security, closed-feedback loop AI can be developed to provide a deep, interconnected understanding of the IT-OT enabled healthcare systems, and to autonomously monitor, analyze and adapt to threats and contexts in Critical Infrastructures (CIs). A closed-feedback loop AI typically refers to a system where the output or actions of an AI model influence and potentially modify the input data for future iterations [2]. This feedback loop allows the AI system to learn and improve over time based on the outcomes of its own actions, which is powerful for self-learning to adapt to dynamic environments and improve their performance based on real-world outcomes. Cognitive Digital Twins (CDTs) presents a promising approach to develop the closed feedback loop AI for monitoring, predicting, responding to and recovering from cyber threats based on virtual simulation, data-driven capability, and behavioral and impact analysis.

We have conducted preliminary research on CDT that is self-learning and proactive, combining Digital Twins (DTs) and cognitive capabilities to interpret and predict unforeseen security issues of the physical system [3]. In [4], we proposed a three-stage framework for enhancing cyber security in healthcare using DT technology consisting of three key modules: (a) physical-twin, (b) cyber-twin, and (c) cyber security automation. In [5], we highlighted the importance of cognitive architecture for simulating human cognitive behavior in cyber security for monitoring, analyzing, and responding to changing security threats to Cyber Physical Systems (CPS). Later, we applied the developed cognitive architecture for developing a CDT for healthcare applications [3]. The past experience in [5] led to the realization that human cognitive capabilities have a huge potential to understand the behavior of digital and physical twins and can be embedded into support decision-making in real systems. Addressing the diverse cyber security threats in the changing environment through a dynamic CDT solution can also be useful for training professionals and improving system performance in IT-OT healthcare.

1.2 Our Contributions

In this paper, we proposed a conceptual CDT approach that mainly includes developing physical and virtual twins, real-time synchronization between them, knowledge base for security and privacy events from IT and OT healthcare systems, and cognitive healthcare decision-making loops. The cognitive cycle in the proposed conceptual framework can be primarily used for simulating human cognitive behavior to automate deployment of cyber security measures against evolving cyber threats. Our approach involves closed-feedback loop techniques between the physical healthcare twin and virtual healthcare twin for investigating complex behaviors of different healthcare stakeholders and systems involved and adapting to security variations in the cognitive process of IT and OT healthcare systems.

1.3 Structure of the Paper

The rest of this paper is structured as follows: Section 2 describes the related work, followed by the proposed framework in Sect. 3. Section 4 presents different use cases and scenarios that could utilize the proposed framework. Finally, Sect. 5 presents conclusions and future work directions.

2 Related Work

This section reviews key studies and developments within (cognitive) DTs in cyber security, highlighting how our work addresses some of the open challenges and extends existing knowledge.

A well-known issue is the slow and reactive nature of current cyber defense mechanisms. Nguyen proposed the Cybonto conceptual framework and ontology to address this issue [6]. This framework aims to utilize DTs and Human Digital Twins (HDTs) to simulate and understand adversaries’ behaviors and tactics for proactive cyber defense. The Cybonto ontology, built on psychological theories and network centrality algorithms, documents constructs and cognitive paths to enhance digital cognitive architectures. Nguyen’s use of DTs and HDTs represents a new approach to proactive defense strategies and extends the current digital cognitive architectures; however, the demonstration of the framework is lacking.

Pirbhulal et al. discusses a novel way to improve security in healthcare systems by employing DTs with IoTs [4]. This approach could help predict and identify potential cyber-attacks/threats, thereby enhancing the protection of patient safety and healthcare services seamlessly in real-time. Moreover, Pirbhulal et al. suggests that DTs can play a crucial role in cyber security by offering insights into potential attacks, identifying system weaknesses, and helping develop better defense mechanisms. They also point out that while the use of DTs has many benefits for cyber security, more applied research with experiments and simulations are necessary for reliable applications in healthcare. In addition, Pirbhulal et al. presents a novel CDT architecture aimed at enhancing cyber security in IoT-based smart homes for telemedicine [3]. The approach focuses on dynamic threat detection and mitigation through continuous monitoring, analytics, and simulation within a cyber twin. Their work also emphasizes the use of AI and ML for analyzing complex behaviors and adapting to evolving security threats and the environment.

Other studies examine DT applications in different domains including smart cities and maritime. Sabeur et al. detailed the S4AllCities project’s development of three distinct DTs for enhancing cyber and physical security of smart urban spaces [7]. This project utilizes distributed edge computing IoT, malicious actions information detection system, and augmented context management system under a system of systems architecture. It focuses on real-time observations and data processing for intelligent monitoring and threat detection, incorporating advanced AI techniques and data fusion frameworks for improved situational awareness in urban environments. Xu et al. introduces a novel framework named LATTICE, which incorporates curriculum learning into the DT-based anomaly detection method ATTAIN to optimize its learning paradigm [8]. This approach is designed to improve anomaly detection in CPS by assigning difficulty scores to data samples, thus enabling the model to learn from easy to difficult samples. LATTICE has been evaluated using publicly available datasets from CPS testbeds, demonstrating its superiority over ATTAIN and other state-of-the-art anomaly detectors in terms of F1 score on model performance and training efficiency, while maintaining competitive detection delay times. Epiphaniou et al. integrated cyber modeling and simulation with DTs and threat characterization for security assessments in IoT and CPS [9]. They presented a comprehensive study on cyber resilience testing, detailing methodologies for integrating cyber standards and simulation standards to address IoT/CPS vulnerabilities. Their work also includes a case study on the Port of Southampton, demonstrating how DTs can simulate cyber-attack scenarios and improve resilience. Epiphaniou et al. contributed to cyber resilience in CIs with a novel approach that combines DTs with AI-enabled threat characterization and cyber modeling and simulation standards.

Finally, Eman Shaikh et al. investigated the security of DTs, which is also essential in applications for safety-critical domains like healthcare [10]. Eman Shaikh et al. presents a comprehensive framework for assessing the security of DTs in various domains, such as healthcare, using probabilistic model checking [10]. Their framework incorporates multi-layered security analysis, including both attack and defense mechanisms across physical, communication, virtual, and application layers of DT systems. Their study also emphasizes the use of Markov Decision Processes and Discrete-Time Markov Chains to model and analyze the security properties and attack probabilities within these systems. One of the key contributions of their work includes a detailed case study on in-patient monitoring systems in the healthcare sector, demonstrating the framework’s application to evaluate the likelihood of successful attacks and the effectiveness of defense strategies. Due to the several advantages of DT technology in healthcare sectors, it has been used in different medical applications such as remote healthcare [11], thoracic healthcare monitoring and diagnosis [12], digitalization in healthcare [13], early mental illness detection [14], personalized therapeutics and pharmaceutical manufacturing [15], inpatients’ falls risk management [16], resilient patient-centered healthcare services [17].

In summary, the literature contains studies that address both the use of DTs in cyber security for different domains in addition to dealing with cyber security of DTs. The main applications of DTs for cyber security included anomaly detection, cyber resilience testing. However, there are several open issues, which we aim to address in this work: the aspect of IT-OT convergence, which is becoming more prominent in CPS, and the essential feature of human-in-the-loop for informed and improved decision-making based on the outputs from DTs. Finally, other essential use cases are missing, like ensuring data privacy and compliance while using robots to assist patients, through DTs in safety-critical application domains like healthcare.

3 Proposed Framework

In this study, we develop a CDT-enabled approach for securing healthcare systems. This can provide secure communication between patients and healthcare providers, which in turn ensures end-end security of critical health applications in addition to data privacy and compliance, as shown in Fig. 1. A CDT represents a sophisticated blend of DT with AI and ML capabilities including human-in-the-loop approach, creating dynamic virtual replicas that can learn, adapt and make informed decisions. Unlike traditional DTs, CDTs incorporate cognitive functions that enable learning, adaptation, and improved decision making. The developed framework consists of mainly three parts: (i) physical healthcare twin, (ii) virtual healthcare twin, and (iii) cognitive healthcare decision making cycle as shown in Fig. 1.

Fig. 1.
figure 1

Illustration of Proposed Conceptual Cognitive Digital Twin-based Adaptive Cyber Security Framework for IT-OT enabled Healthcare.

Full size image

Physical Healthcare Twin.

The first part is an IT-OT based physical healthcare system, in which different stakeholders such as patients, doctors, emergency services, hospitals share sensitive medical information via 5G and the internet. For instance, in remote healthcare infrastructures, both IT and OT systems are used to facilitate remote services. Efficient patient data management through Electronic Health Records (EHRs) [18] represent an example of IT systems. On the other hand, the OT part of this infrastructure could be wearable sensors used to monitor patient’s health, acquire, and transmit data to the hospital in addition to actuators like an airbag in a fall detector [19]. However, integrating IT and OT enables healthcare providers to offer services like telehealthcare more effectively. Since IT and OT systems in healthcare have different predefined capabilities, features and priorities, integrating these two creates new vulnerabilities, making healthcare systems susceptible to cyber-attacks and evolving threat landscape. In healthcare and other safety-critical domains, it is almost impossible to perform cyber security testing on real infrastructures as sensitive healthcare services such as patient monitoring, telemedicine, remote surgery needs to be constantly available and cannot be halted. Thus, cyber twins of physical healthcare systems can be useful in such situations. For instance, analyzing massive volumes of data from IT and OT systems of healthcare systems to identify patterns and anomalies, which may indicate potential ongoing cyber-attacks and also to some extent automate certain cyber security tasks.

Virtual Healthcare Twin.

The second part of this framework is the dynamic virtual representation of the physical healthcare components, systems, and processes. This is mainly in the digital world with real mapping to their entire lifecycle by utilizing physical information, virtual information, and interaction between both sources. Existing studies have developed DTs for cyber security, mainly anomaly detection, in addition to healthcare applications [20, 21]. They are primarily focused on how to use DT for either IT or OT healthcare security, adding cognitive features to DT in addition to the knowledge base on security and privacy events for IT-OT integration in healthcare are the key highlights of our framework. The knowledge base in our framework on security and privacy events is a catalog of different scenarios. It comprises security and privacy events involving IT-OT systems from both publicly available information, data from simulations and experiments performed in the virtual healthcare twin, and inputs from the human involved. One of the key elements of this framework is real-time synchronization of physical and virtual twins. For instance, as soon as any security and privacy risks are identified either from IT and OT healthcare systems in the physical world, it will be communicated and simulated in the virtual counterpart. This is mainly for enhancing the security and resilience of IT and OT systems by providing predictive analytics and actionable intelligence on targeted and effective response actions.

The virtual healthcare twin is also linked with a cognitive healthcare decision making cycle to utilize human-in-the-loop approach, which will be detailed in the next part. This can be useful to improve the security of IT and OT systems by identifying potential vulnerabilities and providing real-time security and privacy events detection and response by receiving inputs and/or validating the diagnosis performed by the decision support via human-in-the-loop, which is essential.

Cognitive Healthcare Decision-Making Cycle.

The third part is the cognitive healthcare decision making cycle, which includes the cognitive model based on Observe, Orient, Decision, Act (OODA) loop concept [22], decision support and computational tools using AI/ML, and feedback loops for continuous learning. This allows dynamically perceiving the IT-OT healthcare circumstances, individuals and social environment behaviors considering the human-in-the-loop approach. This cognitive model has four main stages: learn, plan, observe & orient, and decide & act [5, 25] for continuously collecting, analyzing, and predicting security-related information and medical information to anticipate potential cyber incidents/data breaches and provision appropriate mitigation and/or response measures. The proposed framework applies the feedback loop of decision-making cycle with the physical healthcare twin and virtual healthcare twin to efficiently improve security. This feature will allow the physical healthcare twin to have human-in-the-loop inputs, and also offer human insights while simulating and experimenting with security events in the virtual world.

4 Use Cases and Scenarios

In this section, the different use cases and scenarios on the use of our proposed framework in the healthcare domain are discussed from smart healthcare to remote monitoring and use of social robots for patient support to enable resilient societies.

4.1 Personalized Patient Care and Treatment

In general, the developed framework can be useful for helping patients in understanding different care options, assisting with personal data ownership, secure personalized treatment, and informed decision-making on specific treatments. Current healthcare systems lack efficient ways for testing treatments in advance, but our proposed framework also has the potential to safely enable doctors for testing different treatment options on virtual twins of patient’s before using it. This will allow healthcare service providers to acquire an overview of how the individual will respond to the specific treatment. The use of CDT can allow all these capabilities. For example, with our framework, it is possible to develop secure virtual twins of patients, doctors can simulate the impact of various chemotherapy regimens for a cancer patient, allowing them to select the best possible option for the patients.

Moreover, the existing approaches also lack surgery planning, personalized medicine, disease modeling, and epidemic management since privacy is one of the major concerns. For instance, this is impeding the growth of personalized medicine as the treatment plan is developed through their personal health information, which in this context requires data privacy of patient’s to be handled in a better way. Through our developed framework, this issue can be addressed to some extent with the help of knowledge base on security and privacy events. To that end, we can utilize the privacy scenarios in the knowledge base in addition to the inputs from the human-in-the-loop for identifying potential privacy risks and their impact in personalized patient care through virtual healthcare twins in advance. This can be done through virtual simulation and experimentation to identify best possible actions before deployment.

4.2 Remote Patient Monitoring

Another important use case for our developed framework is secure real-time monitoring of patients. In healthcare, remote monitoring is essential in some situations, where patients need continuous care and support. Some examples of remote monitoring applications for patients that have chronical illness, mental health issues, diabetes, sleep apnea in addition to accidents. One of the main challenges with remote healthcare monitoring is security, because attackers might eavesdrop on real-time communication using potential vulnerabilities in the system. Thus, our developed framework can be used to identify and respond to the cyber security risks. The virtual accessibility of healthcare can provide better understanding of IT-OT cyber-attacks and risks and facilitates the detection and prediction of intentional and unintentional risks.

For example, a flooding attack is one of the potential attacks in remote patient monitoring. This can make it hard for the healthcare providers to identify the potential cause that has compromised the availability of a specific medical device in the infrastructure. In such scenarios, our proposed framework can support identifying the potential cause in real-time, corresponding vulnerabilities (example: vulnerable ports of the devices by performing simulation and experiments in the virtual twin), and how to mitigate it. Human-in-the-loop in our framework can also be used to train and test operators on the remote diagnostic process to identify the extent of the compromise, level of availability and integrity of the control and safety systems and take appropriate actions to respond to a loss of control.

As a part of Adaptive Security for Smart Internet of Things in eHealth (ASSET) project [23], we developed a testbed for monitoring the health data remotely. In [24], we developed the DT infrastructure based on the ASSET testbed [23]. Based on these developments, we intend to apply our proposed framework for improving security of daily activities, end-end monitoring of patients in smart homes. Also, the developed framework will serve as an input for other related activities such as autonomous adaptive security for 5G-enabled IoT systems in Norwegian Centre for Cybersecurity in Critical Sectors (NORCICS) project [25], and dynamic safety and security risk assessment for different critical applications in European Lighthouse to Manifest Trustworthy and Green AI (ENFIELD) project [26].

4.3 Assistive and Social Robots in Care

The third use case is about the applicability of our developed framework in assistive and social robots to offer secure and reliable smart care services. The use of assistive and social robots to support healthcare services are becoming popular, especially for elderly, individuals requiring rehabilitation, and those with disabilities [27, 28]. Social robots in healthcare are designed to interact with people in a personal and engaging manner, providing support, companionship, and assistance in various health-related contexts [29]. As a hospital where robots are used for care services, it is important to ensure the data privacy and compliance of data gathered from different stakeholders for providing personalized support, companionship, and assistance in various health-related contexts. This is to ensure that it complies with laws and regulations like GDPR. Our proposed framework has potential to address these issues since it includes a human-in-the-loop approach for receiving inputs on user willingness to share data and corresponding privacy risks.

Likewise, as a part of the User-centered Security Framework for Social Robots in Public Space (SecuRoPS) project, we use social robots to offer personalized services to citizens/passengers [31, 32]. To that end, it is important to understand users’ willingness to share personal information and which type of personal information they are willing to share with social robots in public space for receiving personalized services. For this purpose, we conducted pilot studies involving potential users’, which is time consuming and cumbersome. However, our developed CDT based framework can play an important role in addressing this issue in the healthcare sector by modeling how social robots collect, store and process data, social robot developers can identify potential data breaches. On the other hand, healthcare workers and managers can use humanoid robots as an assistant in their daily routines [30]. Security is one of the important concerns in such applications as they are dealing with patients in some cases. Moreover, there is no comprehensive overview of different security risks corresponding to it. Therefore, our framework can support in identifying different potential security risks corresponding to such application through simulations and experiments in virtual twins in addition to inputs from humans that can validate the diagnostic.

4.4 Enhancing Resilience in Healthcare Technologies

The developed framework can be useful for critical healthcare applications such as in cancer diagnosis to recognize potential vulnerable functions for increasing cyber resilience. Considering the software complexity of healthcare systems, there is a need for a completely automated solution to support cyber resilience assessments in real-world situations. In [33], it was addressed how to achieve cyber resilience for diagnosing lung cancer using DTs. However, it does not include inputs from both IT and OT systems, also the human-in-the-loop approach was not considered. These factors are important for ensuring resilience in healthcare. Therefore, our proposed conceptual framework includes a catalog based on security and privacy events with inputs from IT-OT systems that can be helpful in enhancing the resilience of healthcare systems by preventing cyber threats and reducing the negative effects or possible recovery to some extent using training and previously acquired knowledge. CDT may improve cyber resilience and build on security, policy, and risk management to provide the trustworthy required for healthcare employees, partners and patients to work with a hospital for embracing digitization efforts.

5 Conclusions and Future Work Directions

As IT and OT become more integrated in healthcare applications, this has increased the potential of adversaries to gain access and even disrupt legacy OT systems due to the emergence of new attack surfaces. Some well-known attacks, including malware, ransomware, can cause operational disruptions and even physical damage in healthcare infrastructures. Therefore, in this study, we developed a conceptual framework for improving security for IT-OT enabled healthcare systems. Important components of this framework include the twin technology, cognitive science, decision support and computational tools, and human-in-the-loop approach. DTs have already been proposed for cyber security and healthcare applications, which lack integrating IT-OT along with knowledge base on security and privacy events and cognitive capabilities. In line with this, the proposed conceptual CDT-based adaptive cyber security framework addresses unique security challenges of IT-OT integration in healthcare. We also presented four different use cases and scenarios on healthcare for potential application of our conceptual framework. Some future work directions of this study are evaluating the applicability and feasibility of the proposed conceptual framework using the suggested use cases and scenarios, identifying opportunities up to which extent proposed framework can be used for improving cyber security training, knowledge transfer and awareness.