1.1.1: H-IoT architecture

The abrupt increase in the usage of wearable devices and fitness trackers over the past few years indicates the exponential increase of these devices and the implants in the future (Statista Research Department, 2016). The integration of smart health monitoring systems with IoT infrastructure has motivated the development of the IoT networks for the healthcare system known as IoThNet. The enormous potential of these systems to track the health conditions of the patient’s vital organs for better diagnosis and medical care has raised the need for the development of standardized architecture. A standard architecture would be the key enabler for H-IoT systems. The IEEE standardization working group is established for point-of-service healthcare devices, which define the communication protocol stack for H-IoT, and it is key parameter indicators (KPIs) (IEEE Standard Association, 2018).

1.1.2: Three-tier H-IoT architecture

The traditional IoT encompasses three basic elements that are hardware sensors, communication enabling technologies, and servers for data processing. These elements form a three-layered H-IoT architecture: (1) things layer, (2) communication layer, and (3) processing layer as depicted in Nauman et al. (2020) (Fig. 1).

Things layer: The first layer of H-IoT architecture is known as things layer. Literature refers this layer as perception, sensor, or device layer. This layer consists of hardware sensors or actuators as things. The hardware sensors record various indices of the H-IoT system based on the application, while the actuator is a feedback system which takes input from user after processing. Sensors transmit the acquired data on uplink transmission, while actuators have downlink transmissions as well for user feedback or instructions. The major objective of this layer is to connect things in the H-IoT network. The things sense and acquire data from the physical world and transmit the data to processing servers via gateways.

Communication layer: The things layer is connected to the processing layer via the communication layer. It is the middle layer, also known as the transmission layer. The layer is virtually divided into two sublayers that are access layer and the core network. The main objective of the access layer is to connect things and applications through gateway or interfaces using communication protocols. The core sublayer determines the optimum route for data transmission. The low-power wireless communication protocols utilized in this layer are Bluetooth Low Energy, Zigbee, Radio Frequency Identification, and Wi-Fi.

Processing layer: The acquired data at the thing layer are processed at the processing layer. The processing layer analyzes the data for extracting useful information that is termed as features using local servers or remote cloud processors. The amount of data generated at things layer is substantial, so the cloud-based solutions for processing are more flexible. However, transmitting all the acquired data to cloud incurs significant delay, which can be reduced using local processing units known as edge node. Sometimes an additional distributed computing layer is included, which is known as fog layer. The additional fog layer reduces latency, improves processing, enhances security, and supports interoperability.

2.1.1: Fitness tracking

Fitness tracking is one of the major applications of H-IoT using electronic wearable devices, which include smart wrist bands and smart clothing (Haghi et al., 2017). The fitness band monitors and records motions and pulse rate, while the smart clothing monitors cardiac activity. The collected data are transmitted to cloud servers using enabling technologies to determine the status of health of the user. The sensor layer is the input interface between the cloud/local server and the application layer (user) in a three-layer architecture. The locally preprocessed data are sent to the cloud database server for storage. The cloud database can be remotely accessed by the doctor or user for monitoring or tracking. The sensors mostly used in fitness tracking include pulse sensors, temperature sensors, and accelerometers. All the sensors are attached to the commonly available fabric that emulates a smart fabric (Kansara et al., 2018). Fig. 2 summarizes the overview of the fitness tracking system.

2.1.2: Neurological disorders

The detection and diagnosis of neurological disorders, such as Parkinson’s disorder (PD), epilepsy, and Alzheimer’s disease, are one of the major application areas of H-IoT systems. The electrical activities of the brain are called as electroencephalogram (EEG). The EEG data are used for neurological disorder diagnostics. The EEG is considered to be the standard tool for neurological disorder diagnosis. The H-IoT systems are utilized to monitor the body temperature, body movements, and audio for epileptic seizure detection (Jagtap and Bhosale, 2018). EEG is used for the detection of epilepsy by mounting sensors on a headband that is connected to an edge node, which also acts as an intermediary node. The gateway processes the data and generates as emergency alert to alert the custodian. The gateway transmits the data to the cloud server for long-term storage and precision analysis by healthcare professionals (Lin et al., 2018).

The major symptoms of PD are tremors. Accelerometer and gyroscope are used to quantify tremors. Sensors are used to record the body movements, and treatment for the patient is determined from the generated and recorded data. The data are preprocessed at the device level and transmitted to the diagnostic level via a gateway. The diagnostic level has an interface, for example, mobile application (Vijay et al., 2018). Freezing of gait is one of the symptoms of PD, and inertial sensors within a smartwatch are used to track vital signs of the body movements. The diagnostics follow the same three-tier architecture that constitutes smartwatch as a device layer, smartphone as a gateway, and cloud servers as processing layer (Šatala et al., 2018). An overview of the architectural framework for seizure suppression system for epilepsy is shown in Fig. 3.

2.1.3: Cardio vascular disorders

Cardiovascular diseases (CVDs) are the types of diseases that affect blood vessels and heart. The most common CVDs include high blood pressure, hypertension, and cerebrovascular diseases, which are referred to as stroke or heart attack. Some of the reasons for heart attack are hypertension, triglyceride levels in the blood, elevated cholesterol, smoking, diabetes, sedentary lifestyle, and obesity. The detection and diagnosis are performed by analyzing the electrical activity of the heart known as electrocardiogram (ECG). The monitoring and analysis of ECG by IoT-based systems are used for the detection and prevention of CVDs. Most of the H-IoT architectures for CVDs follow the same three-tier architecture as shown in Fig. 1. Usually, heart rate and body temperature sensors are used to predict and prevent CVDs.

2.1.4: Ambient-assisted living

The world population is facing a global phenomenon known as population aging. It is predicted that 10% of the population of the Organization for Economic Cooperation and Development (OECD) countries will be more than 80 years old. This will surge the dependency on healthcare facilities exponentially. IoT-based assisted ambient living (AAL) can assist the remote behavior monitoring, emergency detection, and alert generation such as pollution-level alerts (Wan et al., 2017). Wan et al. (2017) summarized the four-layered H-IoT-based AAL architecture as shown in Fig. 4. The sensing layer constitutes the sensors and trackers. The networking layer is composed of communication enabling technologies like Internet, wide area networks (WAN), and personal area networks (PAN). The third layer is the data processing system in the architecture with faculties for multiple approaches. The fourth layer is the application layer, which provides the interface to users for AAL support systems.

2.2.1: QoS improvement

The QoS improvements required for the H-IoT system include low latency, low-power operation, security, and real-time operations. These requirements are explained as follows:

1. 1. Low latency: The time-critical nature of the H-IoT application requires minimum latency. The total end-to-end delay is the sum of transmission delay and processing delay. The transmission delay is reduced by selecting the enabling communication technology with wide bandwidth availability. The processing delay is minimized by utilizing ML-based approaches, fog/edge computing, and cloud computing. The combination of ML, fog/edge, and cloud computing would significantly improve the system performance (Kumar et al., 2018).
2. 2. Low-power operation: The miniature of IoT and wearable devices require minimum power consumption so that the devices should be recharged after a long period of time. The wearable devices can be recharged; however, implants require a battery that can sustain a battery time lasting for years. Therefore, novel and innovation solutions are required for the development of batteries that are long lasting and safe for use within the body (She et al., 2019). In addition, lightweight operating systems (OS) are required to operate with low-power consumption. Moreover, efficient resource management can enhance energy conservation. The efficient utilization of limited power, memory, and processing capability can lead toward the optimum working of H-IoT sensors. ML-enabled energy harvesting algorithms are one of the most suitable solutions to optimize the H-IoT systems (Ortiz et al., 2016). ML algorithms can easily optimize the OS, resource management, and overall performance of the system.
3. 3. Security: The confidentiality of the patient’s data is of paramount importance in H-IoT systems. The manipulation in the user’s data or communication link has serious safety implications. The cryptographic techniques enable secure data access for only authorized users. However, unorthodox methods compromise the patient’s private data. Intelligent security techniques are required to mitigate new attacks. AI provides effective and efficient solutions. However, resource-constrained H-IoT sensors to require lightweight algorithms, so it is imperative to devise efficient ML algorithms (Al-Garadi et al., 2018).
4. 4. Real-time operations: The vast application area of H-IoT includes real-time patient monitoring and teleoperations. Therefore, the substantial amount of H-IoT data should be processed in real time with minimum latency. Extracting useful information refers as features from the data with minimum processing is another challenge. Deep learning (DL) algorithms augment the performance to analyze, process, and extract features with minimum latency and processing.

2.2.2: Scalability challenges

The deployment of the H-IoT system over a large scale in a smart city requires the system to be highly scalable. There are number of factors responsible for scalable H-IoT systems.

1. 1. Scalable deployment: A smart city requires large-scale deployment of H-IoT devices. The same trend is observed with the exponential increase in wearable devices from 80 million in 2015 to 200 million in 2019, which indicates the potential of H-IoT systems (Seneviratne et al., 2017). The scalable and interoperable platforms and underlying communication technologies can enable large-scale deployment of H-IoT devices. Therefore, the standardization of communication technology for heterogeneous wearable and implantable sensors is of great importance.
   In addition, the network resources should be scalable to support large-scale deployment of H-IoT devices. Innovative multiplexing and multiple access solutions for efficient use of the electromagnetic spectrum is in need. The 5G network is expected to support large-scale deployment of IoT devices with an increase in network capacity, while providing 10-fold improvement in energy efficiency. The terahertz (THz) communication is one of the potential solutions for the scarced network spectrum (Chen et al., 2019).
2. 2. Network solutions: The large-scale deployment of H-IoT devices requires efficient network mechanisms. The network should be capable to support massive channel access mechanism in an ultradense environment. The miniature form factor renders the H-IoT devices resource constrained, whereas the channel access mechanism is power-consuming process. Therefore, it is of immense importance to design an intelligent, fast, and low-power consumption channel access mechanism. In this regard, ML provides efficient and promising results. Specifically, reinforcement learning (RL) techniques in ML provide lightweight and distributed algorithms to enhance current standards such as IEEE 802.15.4 and IEEE 802.15.6. In addition, ML techniques can predict traffic patterns in the network and allocate network resources accordingly to meet QoS requirements.
3. 3. Service availability: The H-IoT devices and medical implants are placed on the human body, which is in constant mobility. Due to mobility, the network performance degrades. Hence, service availability and localization are major challenges in mobility. The service must be available in spite of the human mobility. The Internet Protocol version 6 (IPv6) provides an effective solution for network service availability with minimum handover time between different networks. ML algorithms learn the mobility patterns of the network to provide a potential solution for the dynamic network.
4. 4. Interoperability: The H-IoT systems are envisioned to be deployed over a large scale from many original equipment manufacturer (OEM). The data generated from different OEMs vary significantly in format. The interoperability requires data handling, network management, and security. The regulatory authorized needs to put forward an unified standard and data format for heterogeneous devices connected over the Internet.

4.1.1: Real-time monitoring and treatment

The primary purpose that the IoT devices perform is monitoring. However, in the case of healthcare applications, monitoring the patient’s health is equipped with a response system. In addition to the generation of an alarm and alerting the healthcare support staff, the autonomous system can administer a “first aid.” Multiple-use cases can be explored such as:

• • Autonomous blood sugar regulation system: These system monitors and predicts the usage of glucose or blood sugar. ML algorithms can be trained to identify the optimum level of blood sugar and thus automatically administer a suitable level of glucose via an implant.
• • Precision medicine systems: The overdose of prescription drugs is a significant problem in the healthcare industry. Sometimes, the doctors prescribe a dosage higher than the required amount. Therefore, to optimize the administration of drugs, an ML-based precision drug administration module can be implemented. The levels of various chemicals in the bloodstream are continuously monitored, and the user is alerted about the exact amount of the drug dosage. This ML-based system controls the overdose of medicines by training from the standard blood composition. If an anomaly is detected, the drug usage is regulated accordingly.
• • Prediction of neurological and cardiological events: The prevalence of cardiological and neurological disorders, like stroke and epilepsy, highlights the importance of H-IoT in health monitoring. The occurrence of stroke and epileptical seizure is possible to predict from the events on the ECG and EEG, respectively. AI-based algorithms can identify the events or changes on the wave forms that precede a seizure or a stroke. In addition, countermeasures and alerts can be generated before the event occurs. Multiple countermeasures are defined for a stroke or seizure that can be implemented to mitigate the fatal risks of these two disorders.

4.1.2: Training for professionals

The healthcare professionals can improve their skills by using AI-based training modules that can test the knowledge and effectiveness of the medical students by generating random scenarios. These scenarios are based on real-life cases using which the algorithm is trained.

4.1.3: Advanced prosthetics

A large part of the world population faces some form of physical disability, and many among those depends on others for necessary activities. Therefore, ML-based prosthetics work by analyzing the signals in the central nervous system. The analysis can allow a robotic prosthetic package to obey the command of the user.

4.2.1: Channel access

RL algorithms are allowing the use of AI systems in channel access, which has to abide by a strict rule in terms of time delay and reliability. The RL algorithms can form policy and evaluate it to optimize the access to an already limited available channel bandwidth. The random access management can also be improved by utilizing ML algorithms to assign priorities to nodes and data types.

4.2.2: Dynamic data management

The continuous streams of data generated by the sensors are mostly composed of redundant data. Therefore, network resources are unnecessarily burdened. The lightweight and efficient ML algorithms can remove the redundancies in the data and prioritize the events that are urgent. The precious channel bandwidth can be preserved using this approach.

4.2.3: Fully autonomous operation

In an attempt to optimize network management, AI algorithms can be utilized. The use of lightweight ML algorithms is supported by the resource-constrained IoT nodes and network gateways. The split-second decisions can be made autonomously by ML algorithms that learn the traffic patterns and traffic load. The routing algorithms can be improved by predicting the future requirements of nodes based on the past experiences of the nodes.

4.2.4: Security

The security of patient data is a critical part of the H-IoT system. Multiple approaches to protect the privacy and the patient data are in use. However, the use of ML-based algorithms is also being popularized, such as in the case of intrusion detection systems (IDS). The ML algorithms learn the traffic features and patterns to identify a baseline upon which classification algorithms are applied. This approach is extended to other attack types, and the suitable ML algorithm is required for such instances. A randomized controlled behavior can be implemented for controlling the access of the nodes to enhance the system performance by utilizing ML algorithms.