11 Healthcare IoT
541
DES, Blowfish, and Skipjack and asymmetric public-key algorithms such as
the Rabin’s Scheme, NtruEncrypt, and Elliptic Curve Cryptography. As a rule,
symmetric algorithms are less computationally intensive, making them for lowpower 8-bit/16-bit IoT devices. At the same time, problematic key exchange
mechanisms and confidentiality issues often create difficulties [1, 9].
3. Cloud layer: A large body of literature exists on the security issues involved in
the deployment of cloud applications. Any provider of e-Health products and
services will need to establish an efficient, effective set of tactics for proactively
combating the negative impacts of attacks. Widespread vulnerabilities in the
cloud include Denial-of-service (DoS) attacks, SQL injections, malicious code
injections, Spear-Phishing attacks, sniffing attacks, path traversals, unrestricted
file uploading (remote code execution), cross-site scripting (XSS), Trojan horses,
viruses, and brute-force attacks using weak password recovery methods [1, 9].
4. Human layer: The fundamental principle of IoT e-Health security is that
individuals should receive training on how and when to avoid disclosing private
healthcare information. If a knowledgeable group of attackers gain physical
access to an end user’s IoT e-Health device, those attackers could directly pull
data from the device’s internal memory and firmware and modify its settings to
obtain partial or complete control over it. In addition, it will be crucial to train
users to avoid common security pitfalls such as sharing physical or electronic
keys, choosing weak passwords, or purchasing used medical equipment [1, 9].
11.7 Case Study: Collaborative Machine Learning-Driven
Healthcare Internet of Things
To justify the proposed multi-layer architecture for healthcare, as a case study, we
discuss how an online arrhythmia detection can be mapped to the architecture.
In this case study, we build a collaborate solution that distributes the intelligence
between endpoint IoT health device, fog node, and the cloud. This strategy
enables us to have the best of different worlds and to have a tradeoff between
accuracy, communication latency (transmission time), processing time, and energy
consumption. Note that in the cloud we can execute very complex algorithms,
but it has two drawbacks. First, cloud-based arrhythmia detection is not real time.
Second, the device needs to transfer a large amount of raw data (ECG signals) to
the cloud resulting in high power consumption which is not suitable for wearable
devices. On the other hand, endpoint IoT devices and fog nodes can provide realtime decision making; however, unfortunately we cannot execute holistic machine
learning techniques on those devices due to their limited processing capabilities.
Thereby, this can result in inaccuracy. To tackle this issue, we use a collaborative
technique [19]. In this collaborative solution, we distribute the intelligence across
device, edge, and the cloud. We utilize a shallow neural network to fast ECG-based
anomaly detection in the endpoint IoT devices. Although the accuracy of this type
541
DES, Blowfish, and Skipjack and asymmetric public-key algorithms such as
the Rabin’s Scheme, NtruEncrypt, and Elliptic Curve Cryptography. As a rule,
symmetric algorithms are less computationally intensive, making them for lowpower 8-bit/16-bit IoT devices. At the same time, problematic key exchange
mechanisms and confidentiality issues often create difficulties [1, 9].
3. Cloud layer: A large body of literature exists on the security issues involved in
the deployment of cloud applications. Any provider of e-Health products and
services will need to establish an efficient, effective set of tactics for proactively
combating the negative impacts of attacks. Widespread vulnerabilities in the
cloud include Denial-of-service (DoS) attacks, SQL injections, malicious code
injections, Spear-Phishing attacks, sniffing attacks, path traversals, unrestricted
file uploading (remote code execution), cross-site scripting (XSS), Trojan horses,
viruses, and brute-force attacks using weak password recovery methods [1, 9].
4. Human layer: The fundamental principle of IoT e-Health security is that
individuals should receive training on how and when to avoid disclosing private
healthcare information. If a knowledgeable group of attackers gain physical
access to an end user’s IoT e-Health device, those attackers could directly pull
data from the device’s internal memory and firmware and modify its settings to
obtain partial or complete control over it. In addition, it will be crucial to train
users to avoid common security pitfalls such as sharing physical or electronic
keys, choosing weak passwords, or purchasing used medical equipment [1, 9].
11.7 Case Study: Collaborative Machine Learning-Driven
Healthcare Internet of Things
To justify the proposed multi-layer architecture for healthcare, as a case study, we
discuss how an online arrhythmia detection can be mapped to the architecture.
In this case study, we build a collaborate solution that distributes the intelligence
between endpoint IoT health device, fog node, and the cloud. This strategy
enables us to have the best of different worlds and to have a tradeoff between
accuracy, communication latency (transmission time), processing time, and energy
consumption. Note that in the cloud we can execute very complex algorithms,
but it has two drawbacks. First, cloud-based arrhythmia detection is not real time.
Second, the device needs to transfer a large amount of raw data (ECG signals) to
the cloud resulting in high power consumption which is not suitable for wearable
devices. On the other hand, endpoint IoT devices and fog nodes can provide realtime decision making; however, unfortunately we cannot execute holistic machine
learning techniques on those devices due to their limited processing capabilities.
Thereby, this can result in inaccuracy. To tackle this issue, we use a collaborative
technique [19]. In this collaborative solution, we distribute the intelligence across
device, edge, and the cloud. We utilize a shallow neural network to fast ECG-based
anomaly detection in the endpoint IoT devices. Although the accuracy of this type