15.2.1 Smart Sensors: Intelligent, Biocompatible, and Wearable

Sensing technology has had rapid growth in the past decade. For example, in ophthalmology, retinal implants, as multicolour light sensors, have made a drastic transformation in restoring vision (although, further research is required to form a benchmark) [1]. A flexible wireless electrocardiographic (ECG) sensor with a fully functional microcontroller has been developed by IMEC in Netherland. An e-textile system for remote, continuous monitoring of physiological and movement data are offered by Smartex, Italy. In this garment, the embedded sensors allow both ECG and electromyographic (EMG) data capturing. Additional sensors facilitate thoracic and abdominal signals data recording associated with respiration and movement. Smartphone-based ECG monitoring system has been commercialised by IMEC. The ProeTEX project by Smartex aims to develop smart garments for emergency responders. Researchers in the University of Houston, USA have reported the design of a multifunctional ultra-thin wearable electronic device, a mechanically imperceptible, and stretchable human–machine interface (HMI) device, worn on human skin to capture multiple physical data or worn on a robot to offer intelligent feedback, forming a closed-loop HMI. To alleviate the difficulties in wearing electroencephalography (EEG) sensor cap, some electrodes have been embedded in the glass handles in a recently introduced EEGlass [2].

Smart sensors receive their inputs from the physical environment and based on the data decide on what to do next. This can happen mainly due to the built-in computational resources which perform predefined functions. These sensors enable an automated and more accurate collection of environmental data with less error, more accuracy, high robustness, and perform self-diagnosis for human body and self-calibrating within the BSN. Such devices have onboard processing which can analyse the captured data and make decisions. Finally, as in many other sensors, they can communicate with other devices using wireless technology. Additionally, by incorporating artificial intelligence with the sensor technology, a new generation of sensors will be able to learn from the environment and the data. Moreover, smart sensors will promote a new generation of cooperative and decentralised (as opposed to centralised) BSN networks. These networks are central to future Internet-of-things (IoT) technology. Adaptive cooperative networks and cooperative learning are the two advanced techniques in signal processing and machine learning which focus on decentralised distributive systems.

There are enormous applications for smart sensors. For example, in wound care, it is crucial to keep the wound cover untouched unless its exposure is imperative. This reduces interruption to the wound healing process. To enable this, the sensors can be embedded within the bandage to constantly check the state of the wound and decide whether there is need for a change or for medication. Another example is the sensor used for food checking in supermarkets. Other smart sensors include smart needles to analyse and identify blood vessels or tissues in brain surgery which involves a great deal of signal and image processing; iTBra, for the detection of breast cancer by recognising the thermal abnormalities or tissue elasticity; the CADence System, for the detection of coronary artery blockages using cardiac sounds; UroSense, for urine monitoring with the help of core body temperature (CBT) and urine measures; Digital Pill, which senses electric signals in the stomach fluid to prescribe suitable medication [3]; and as another interesting device, a sociometric badge which is a wearable computing platform for measuring and analysing human behaviour, based on variability in a person's speech spectrum, in organisational settings [4]. Integration of these sensors within a BSN will certainly boost the effectiveness and inclusiveness of such networks for better human body monitoring.

Biocompatibility of sensors measuring internal body, biological, or metabolic activities is another issue in sensor technology. Some recent developments include biocompatible soft fluidic strain and force sensors for wearable devices [5]. This is a silicone-based strain and force sensor composed of potassium iodide and glycerol solution. Currently, a considerable amount of research on designing biocompatible sensors and devices has been undertaken. Recently, Harvard University researchers developed a soft nontoxic wearable sensor that unobtrusively attaches to the hand to measure the force of a grasp and the motion of the hand and fingers [6]. The sensing solution is made from potassium iodide, which is a common dietary supplement, and glycerol, which is a common food additive. Nontoxicity and highly conductive liquid solutions are two novel elements of this sensor. After a short mixing duration, glycerol breaks the crystal structure of potassium iodide, resulting in potassium cations and iodide ions. This makes the formed liquid conductive. Since glycerol has a lower evaporation rate compared to water, and the potassium iodide is highly soluble, the liquid is both highly conductive and stable across a range of temperatures and humidity levels. Similar to smart sensors, designing and using biocompatible devices can make a significant contribution to more effective BSN designs. This, to a large extent, depends on the replacement of traditional components with new ones made of polymers. Chinese researchers have developed a stretchable biocompatible metal-polymer conductor that has potential for wearable electronic circuits bridging electronics and biology [7]. Printable, highly stretchable, and biocompatible metal-polymer conductors have been introduced by casting and peeling off polymers from patterned liquid metal particles, forming surface-embedded metal in polymeric hosts. These products are the samples from the incredible world of biodegradable and biocompatible polymers for sensor designs [8].

The adoption of wearable sensors is one outstanding problem for BSNs. The sensor's measurement not only has a valuable impact on patient monitoring but also can be used by occupational safety and health professionals, athletes, and computer game players. Therefore, there are potential benefits of using such technologies in the workplace. Nevertheless, there are perceived barriers which prevent the widespread adoption of wearable sensors for most of these applications, particularly in industrial workplaces. Thus, many workplaces are hesitant to adopt these technologies.

In a survey of 952 valid responses from public, health, and safety professionals, over half of the respondents were in favour of using wearable sensors at their respective workplaces [9]. Nevertheless, concerns regarding employee privacy/confidentiality of collected information, employee compliance, sensor durability, the trade-off between cost and benefit of using wearables, and good manufacturing practice requirements were barriers and described as challenges hindering adoption. Based on this study, it was concluded that the broad adoption of wearable technologies appears to depend largely on the scientific community's ability to successfully address the identified barriers. For general applications of wearable sensor systems, several approaches may be followed. These include: (i) design of more miniaturised wearables, (ii) make less intrusive sensors, (iii) improve sensors appearances, (iv) use lighter materials, (v) embed it within current multimedia gadgets such as wireless headphones, and (vi) more closely follow the needs of different users.

15.2.2 Big Data Problem

Identifying the health status of the human body is more accurate when there is access to many biometrics over a long duration. A combination of a large amount of data recorded using various modalities is valuable since there currently exist powerful mining, archiving, learning, and processing tools and algorithms for analysis of such data. The term ‘big data’ does not necessarily refer to the size of data; instead, it implies that the data cannot be processed in a single core computer or processing unit. Therefore, one aspect of future work on big data is to develop signal processing and machine learning techniques that can be run in a distributive manner over multicore computers or computing clusters [10].

15.2.3 Data Processing and Machine Learning

The underlying information derived from raw data is often very rich. Nevertheless, powerful signal processing algorithms have to be designed and implemented to uncover and extract this information. Current techniques in both diffusion (such as tensor factorisation or diffusion adaptive filters) and fusion (such as various data mining algorithms and transcription methods) can further be improved or regularised by incorporating subject (such as peripheral clinical and personal information), data (such as the nature of data, their dimension, colour, sparsity, smoothness, and variation boundaries), and environmental (such as temperature, system limitations, and noise) constraints into the algorithms.

The availability, multimodality, multidimensionality, and multiclass nature of data and computation capacity of the new machines make deep neural networks (DNNs) outperform other classifiers. While these classifiers still require very high computational power, the number of DNN architectures is also ascending rapidly. One new direction in DNN design is a deeper analysis of natural brain network functioning. One example of using this concept is by means of biological learning algorithms [11, 12]. These systems present a theoretical framework for understanding the regularity of the brain's perceptions, its reactions to sensory stimuli, and its control of movements. They offer an account of perception as the combination of prediction and observation: the brain develops internal models that describe what will/should happen next and then combines this prediction with reports from the sensory system to form a belief. The biological learning algorithms outperform the optimal scaling of the learning curve in a traditional perceptron. It also results in a considerable robustness to the disparity between weights of two networks with very similar outputs in biological supervised learning scenarios. The simulation results indicate the ability of neurobiological mechanisms and open opportunities for developing a superior class of deep learning algorithms. Since the new direction in machine learning system design is to sense and learn from the data then the more data available from a diverse range of sensors, the more precisely the state of human body can be described or identified.

15.2.4 Decentralised and Cooperative Networks

As soon as smart sensors become popular, there will be a shift from centralised systems (including a hub and a number of sensors directly connected to it) to a decentralised network, where there will be no central hub and the body sensors can process the data, cooperate with each other, make decisions, and also communicate with other sensors, including those outside the BSN. To facilitate this, new cooperative models through consensus and diffusion adaptation techniques should be developed which consider each sensor as an intelligent agent. Advances in graph theory and related signal processing techniques may be employed to best model the connectivity among the body and environmental sensors. Adaptive cooperative algorithms should be able to model the interactions between two or more BSNs and also between a BSN and an external wireless sensor network (WSN). As an example, consider a patient suffering from asthma, equipped with a wireless body area network (WBAN), walking in an area with a variable level of air pollution equipped with a WSN. A cooperative decision-making system can help this patient find their way through the less polluted regions.

15.2.5 Personalised Medicine Through Personalised Technology

In the context of human monitoring using wearable technology, ‘personalised medicine’ refers to linking an individual's internal (physiological or metabolic) and external (movement in a parametrised environment) bodily status to their monitoring system. Height, weight, diet, living style, amount of sport, a patient's daily activities, the environment they interact with, etc., can be used in a more comprehensive treatment plan. The smart sensors may then be regulated so that the wearable system best reveals the necessary diagnostic information through better and more accurate parameters setting. There is great advantage in such adoption for sport when, for example, different-sized athletes engage in completely different types of sport. Moving one step further, consider the advances and progresses in machine learning and signal processing from predictive models (autoregressive, singular spectrum analysis) to recurrent neural networks, and, very recently, long short-term modelling (LSTM) deep networks. Recurrent neural networks and LSTM exploit the history and background activities of the subject's data to assess their current and future status. Thus, an effective personalised medicine can benefit from these advancements in technology.

15.2.6 Fitting BSN to 4G and 5G Communication Systems

Current mobile systems adhere to long-term evolution (LTE) and 4G technology standards, while 5G is being offered by many communication system developers as a near-future mobile communication system. 5G systems are expected to provide bandwidth efficiency and higher accessibility by providing more intelligent communications and data retrievals. Although the body area network (BAN) short-range communication remains the same but for the data to be available over the World Wide Web as well as mobile systems, the BAN needs to be updated to meet the new generation of communication systems. Energy efficiency, interference mitigation, and wireless power transfer capability are probably the most important factors for making a WBAN suitable for integration within a 5G network [13]. More importantly, as mentioned in Section 15.2.4, in future decentralised systems the smart sensors will be able to communicate with devices outside the BAN local network individually. In such an inevitable scenario, the sensors have to be equipped with full Internet connection and comply with the new generation communication system standards and protocols.

15.2.7 Emerging Assistive Technology Applications

The most popular and effective BAN application is for assistive technology. Both rehabilitative assessments and patient rehabilitation should expand to cover a wider needy population. Vulnerable epileptic, stroke, dementia, depression, paralysis, and many other groups of patients should benefit from the outcome of this research direction now rather than tomorrow. Smart chair, assistive robot, fall detector, driver fatigue detector, hazard detector, and many other devices used by the military for life protection are only a few examples of assistive technology made available to the public.

Assistive technology covers a wide range of products and equipment furnished with assistive robots as the most advanced, effective, and intelligent ones. In the near future, these robots may be able to assist humans both physically and mentally, and offer patients a much better quality of life at home and at work.

15.2.8 Solving Problems with Energy Harvesting

Although new computerised systems are well equipped with low-power processors, in many applications where on-board processing is essential the sensors have serious demand for green energy harvesting. The conversion of thermal, mechanical, and chemical energy to electricity is always a prime objective when choosing the sensors. New systems should also allow energy to be delivered from sensor to sensor as well as the power in the sensor itself or those within its neighbourhood to be predicted. This raises many questions about how the energy can be shared between the sensors or transferred across a sensor network. We may think of a number of solar cell sensors with only a few of them exposed to sunshine for a certain period. Therefore, researchers should constantly look for new sources of energy to better operate the sensors at any required time.

15.2.9 Virtual World

Social affairs and interactions are now far more extensive and complex than before. This requires a more advanced technology to better cater for new demands and user requirements. In addition to a considerable amount of work, mostly in computer game design, using virtual reality (VR) [14], one of these technologies, called augmented reality (AR), which has applications in education, games, business, etc., has gained a prominent position among developers and researchers.

Exploiting AR, the users can see the 2D or 3D digital information (images, text, etc.) at the same time as seeing the rea -world by using a gadget (such as glasses or a cell phone). Through AR, humans interacts with sensor measurements (e.g. the scene captured by a video camera). Using this technology, a blend of augmented and virtual reality system using Holobody technology which, with the help of HoloLens glasses, is able to see the anatomy of the human body in real 3D space have come to practice [15].

Some applications of AR include, but are not limited to, reducing the schizophrenia stigma [16] to improve sufferers' social interactions, fishing training computer games [17], civil engineering education [18], and teaching chemistry for high school/university students [19]. In medicine, this technology is also used to treat disease. In [20] an approach based on AR for the treatment of pain created in the lost limb (phantom pain) is presented. Also, in [21], AR has been used to treat the people suffering from fear from small animals such as beetles and spiders. The results of this research have been quite useful and effective. In stomatology, AR has been suggested for dental treatment and implantology [22].