Integrating smart technologies into the home for assistive purposes has garnered much research over the years. Some, such as MavHome, combine several technologies together to provide a more homogeneous solution whilst others concentrate on one aspect such as patient requirements during the nighttime or Activities of Daily Living recognition.
What follows is a look at several applications of Smart Healthcare systems grouped into a classification based on their technological approach.
3.1. Smart Homes
Although the technologies involved in Smart Homes have evolved rapidly over the last three decades, the principals and techniques have remained surprisingly consistent. A Smart Home is a collection of sensors and actors, and in some cases an intelligence, whose aim is to provide autonomous services to its inhabitants. There are many degrees of intelligence, or “smart”, when it comes to the Smart Home which can range from simple reactive actions based on predefined rules right through to adaptable and self-learning systems able to learn what the occupant needs and act upon it.
For completeness, a look at three key early systems is offered to give an insight into the evolution of the Smart Home. These three projects, all with research published in 1998, show different approaches to information gathering and dissemination: video surveillance and local processing, simple sensors and PSTN network communication, and ambulatory monitoring with automatic emergency recognition. By considering these systems as well as modern approaches a more complete picture can be given.
Microsoft’s EasyLiving project [32
] showcases the early investigations into context aware computing using an array of video capture devices instead of more traditional physical sensors. Using several vision modules in each room, the system can identify motion, people, gestures, and even the surrounding environment including room fixtures and fittings. Individuals are identified by use of an active badge system and data from the video feeds are processed using a distributed computer network. One unique aspect of the EasyLiving project, especially at the time of its creation, was the focus of geometric relationships between people, places, and things. These relationships, or mappings, enable the EasyLiving system to form interaction information that would associate objects with their likely use, which could later be used in a more intelligent system for behaviour prediction.
Another early system targeted specifically at the elderly was developed in the UK by British Telecom (BT) and the Anchor Trust [33
]. To address the ever-increasing demand for community care, the Smart Home system was designed to monitor and report on the lifestyle of its elderly inhabitants. This was achieved using a low-cost sensor network comprised of PIR, door, and temperature sensors. This data was stored along with a timestamp to compile a lifestyle database as a reference pattern. Significant deviations from this pattern, such as long periods of inactivity, are used to detect anomalies whereby an alert message via the telephone system was sent to the patient's designated carer.
], another UK-based system, takes the approach of combining several environmental and personal sensors to provide an after-care system in the home. Centring on the hypothetical scenario of a patient returning home following hospitalisation, CareNet tries to address the aging population and “hospital-at-home” market. A key feature of the system is to provide emergency support by incorporating environmental and ambulatory monitoring with analytical capabilities to facilitate automatic recognition of alarm conditions, taking the onus away from the client to instigate the call for help. CarerNet achieves this using a distributed sensor network and various wearable devices.
Fast-forward 5 years and we begin to see more powerful hardware combined with sophisticated software able to offer more than just reactive functionality. MavHome (Managing an Intelligent Versatile Home) [35
] aims to provide an agent-based rational intelligence able to gather, process, and make decisions upon a multitude of data streams. What makes this system particularly interesting for a healthcare application is its use of a probabilistic model to estimate the inhabitants’ movements. This functionality is constructed by sampling events and comparing them to known sequences or sequence matching. If an event is followed by another that is part of a known chain, there can be a confidence factor associated with the prediction of the next event. The more events occur in sequence, the greater the level of predictability. Once events are detected or predicted, the appropriate action can be instigated. Over time, the system can become more accurate through use of a data-mining algorithm named Episode Discovery, operating on the larger set of data. The work on MavHome was later extended at the Washington State University in a project named CASAS [36
The GatorTech Smart House [37
] from researchers at the University of Florida takes a whole-home approach to provide a “programmable pervasive space”. The team tries to address the problem of closed ecosystems and fixed installations by designing an extensible framework for software and hardware made possible by a generic abstraction middleware. Using a multi-tiered approach incorporating the physical, sensor platform, service, knowledge, context management, and application layers, the system can interpret sensory data and combine it with contextual information about the home, its environment, and its inhabitants to provide assistive living functionality. This “smart house in a box” was aimed at the aging population market providing a system that the “average user can buy, install, and monitor without the aid of engineers”.
Stratogiannis et al. take a different approach, which they term SandS or Social and Smart. They argue that “Filling a home with sensors and controlling devices by a computer is nowadays not only possible, but it is commonly found in homes” [38
]. In their system, users are modelled using persona stereotypes which are a source of information about users and their characteristics. Contextual information is gathered using wireless ambient sensors (temperature, humidity, etc.) and is combined with external smart city data such as weather conditions, people movements, and social interactions. Rules are expressed in a high-level language and knowledge is represented through semantic web technologies such as OWL 2 Web Ontology Language to ensure actor interoperability. The collected contextual information can be used to characterize an entity, i.e., people, place, or thing. Although this system is in its infancy it does shows promise as a flexible and extensible framework for modelling context aware environments.
There are several examples of intelligent smart homes capable of adapting to the occupants needs through a variety of approaches but the system proposed by Pradeep et al., is somewhat unique [39
]. When it comes to more advanced ways of controlling a smart home using brain waves is an approach that not many smart home researchers have investigated let alone produced a prototype to demonstrate its effectiveness. The field of Human Computer Interaction (HCI) has been explored for several decades and often revolves around the use of computer input peripherals and visual/tactile feedback mechanisms but using the Brain Computer Interface (BCI) is relatively new. Table 1
lists some characteristics of the given systems. Columns marked with a ‘●’ indicate that the system does exhibit that particular characteristic.
3.2. Robotic Assistants
In contrast to medical intervention which is primarily given by caregivers, doctors, nurses, and other medical professionals, it is now possible for many day-to-day tasks can be performed by Robotic Assistants. These activities really can make a difference to the health of the patient and can be essential to delivering a holistic approach to care.
Robotic Assistants can perform a whole gamut of operations but often it is the simplest tasks that lead to a patient giving up their independence in the first place. For example, tasks such as forgetting to eat, take medication, or go to the bathroom are real problems that can be addressed by providing intelligent reminders. A wearable device could perform this function but there is a danger that the patient simply forgets to wear the device. Instead, a Robotic Assistant could be aware of exactly where the patient is in the home and proceed to “meet” with them to deliver the reminder.
Robots have often been the antagonist in popular culture made famous by authors such as Isaac Asimov and Philip K. Dick, enough so that many studies have been performed exploring attitudes towards robots using the “Negative Attitudes toward Robots Scale” (NARS). Differing cultural perceptions may play a part in the introduction of robots in the home. Although many reports indicate a positive attitude to the companionship of an electronic aid, a survey of the available case studies involving robots and the elderly shows that a large proportion were undertaken in Japan where attitudes to technology and robotics may not be reflective of other countries. Despite this bias towards one culture Robotics is an area that has merits in the field of home assistance. More research is needed with a broader and more diverse audience, ideally with the same techniques and technologies, to make a more general acceptance statement but as the technology to enable robotic assistance matures the inevitability of their use becomes more apparent.
The biggest barrier to entry today is the prohibitive cost-to-functionality of these devices. Outside of the often single-function robots such as the Roomba (which may also be considered expensive to most), more general-purpose assistive devices are currently the domain of the research community.
3.3. Telemonitoring and Telesupport
There are four key areas of concern in Telemonitoring:
When it comes to vital-sign monitoring data in the Smart Home comes from various devices and sensors placed on or in the patient (in the case of wearables/implantables) or around the home itself (environmental sensors). Often these sensors relay data back to a central hub or gateway that can process the information or send it on to a remote server where it can be combined with patient information for automatic and/or manual interpretation. When dealing with patients that have potential life-threatening conditions such as respiratory congestion or a heart condition, this information can literally be used to save lives through fast medical intervention.
Various systems have been proposed that monitor patient vital signs to provide Real Time Health Advice and Action (ReTiHA) [40
]. One such system, used to monitor paediatric neuroblastoma patients at home, concentrates on a wide range of vital signs including “blood pressure, heart rate, temperature, body weight, C-reactive protein, white blood cell count, wellbeing, pain level, nausea level, skin alterations
” using a smartphone application and a set of sensors [41
]. What is particularly interesting with this solution is the fusion of sensor data with subjective feedback to build an overall picture of the patient’s wellbeing. Using a mobile device (or tablet) the patient is presented with a user interface tailored specifically for children. An Android application named MobileMonitor
collects data from the measurement devices and transmits it to a backend system over the HTTPS protocol for added security. In addition, the Near-Field Communications (NFC) protocol is used in conjunction with an A4 graphical representation of patient wellbeing (good, medium, bad) to capture subjective data. Together, the sensors and wellness data can be used by healthcare professionals that in turn allow the patient to remain at home longer between hospital visits. Alerts are given to the corresponding oncologist if deteriorating conditions are observed.
One of the biggest hazards confronting the elderly at home is the potential of falling. Elderly falls cause around 8 billion dollars of direct medical costs per year in the US alone and are the leading cause of accidental injury and death in the elderly with 68% of all elderly hospitalisations [42
]. This figure increases to 86% in patients over 85 years old. Li et al. propose a wristlet device with accompanying mobile interface for real-time fall detect in the elderly. Current wearables can employ sensors such as accelerometers to detect if a human is falling using either fixed threshold methods, pattern recognition strategies, conventional or fuzzy logic, or artificial neural networks. The problem is that detection accuracy is usually low with traditional software methods leading to medical negligence. In addition, such algorithms often require complex computing resources, draining embedded batteries. Li et al., propose a smart wristlet, giving 24 h of fall detection service with a detection rate of up to 93% with a simplified solution that saves battery life. This can then be coupled with a wearable airbag system to deploy when a fall is detected. The wristlet has a three-layer architectural design: Application layer (software and statistics), Network layer (communications), and Sensing layer (3-axis accelerometer, 3-axis gyroscope, 3-axis compass, and pulse sensor). Sensor data are sent to a mobile phone via Wi-Fi or Bluetooth where they are analysed further. It should be noted that although this research concentrates on fall detection the technology and approach can be extended to heart-related diagnosis, activity monitoring, and even to non-traditional approaches such as those seen in Chinese traditional medicine. The system learns by mining training data to effectively learn what a fall looks like and use this to compare to real-time data while the algorithm used looks at occurrence, or most notably recurrence, where the most relevant data are often seen multiple times, whereas irrelevant data often occur only rarely. Using Term Frequency Inverse Document Frequency (TF-IDF), the process can discard irrelevant data whilst surfacing only the most important features which results in a higher accuracy. Despite some success in a trial of 246 aging people in Beijing (all aged between 59 and 63 years old) this system, similar to others of this kind, does have its limitations. Most notably, ordering and time-sequential problems are present which can be of concern for a real-time critical system like this. The authors of this research are looking at ways of addressing this in the future. Furthermore, battery life of the wristlet is still only 24 h which is limiting for some situations where the user may forget to charge the device or charging is just not possible each day. This is also present in more commercial solutions such as those from Fitbit, Apple, and Garmin, whose tracking devices are all limited by battery life measured in hours rather than weeks.
One of the hottest areas of Smart Home research now is that of determining what a home occupant is doing at a given point in time. Activities of Daily Living (ADL) is the term given to everyday activities that people perform such as washing, eating, and sleeping. For most people these activities form patterns which, when monitored, can be very useful in identifying abnormalities that may be a precursor to patient intervention. Many elderly people, the demographic most susceptible to degenerative diseases, live alone at some point in their lives which puts them at risk not only from their afflictions, but from accidents. According to the most recent Office of National Statistics report on injury and poisoning mortality in England and Wales [43
], published 2013, those over 65 years of age are most at risk, suffering the highest mortality rates and highest accident severity. Falls were the leading cause of accidental death with 30% of male and 39% of female mortalities attributed to this. Another disease that ADL and other Smart Home technologies tries to address is dementia which has seen a threefold increase in mortality in the decade 2004–2014. Overall, people are generally living longer, which poses a challenge for traditional healthcare services.
A list of common ADLs is shown in Table 2
According to research by Ni et al. [44
], there are two major types of ADL: essential basic or personal self-care activities (BADL) such as bathing, grooming, toileting and consuming food and drink, and non-essential instrumented or domestic activities (IADL) such as shopping, watching TV and reading. Additionally, although not formally defined, a third ADL can be extrapolated from the text and defined as ambulatory or movement related activities (AADL) such as walking, running and bike riding. To extend this categorisation we can propose that a fourth ADL group be defined: Social interactions ADL (SADL) that would include activities such as email communication with friends and family, video conferencing with tools such as Skype, telephone calls and in-person visitors. Grouping ADLs into these four categories allows a general wellbeing assessment of the patient rather than concentrating on one aspect such as physical health that provides only a limited view.
Using the extended Ni et al. groups, we can reclassify the ADL activities in Table 2
as shown in Table 3
The actual identification of activities is often difficult as the accuracy of existing indoor tracking sensors and the corresponding analysis software is still in its infancy. Movement tracking, the fundamental foundation of ADL studies, is not always easy to accurately detect, especially in the indoor setting of the home. To overcome this some attempts have been made to combine technologies such as those found in common smart phones such as the iPhone 6S (accelerometer and gyroscope) with software filters to detect occupant movement. In trials, one solution using the software Monte Carlo and Kalman filtering methods managed to get accuracy down to an average 0.47 meter resolution at a 95% confidence level from using a Bluetooth LE sensor network and an off-the-shelf Apple iPhone device [45
]. What makes this research so compelling is the fact that only one form of sensor data was used, which opens the possibility of further improvements by combining sensor technologies. In fact, the authors plan to extend this work by adding other sensor data from force sensitive resistors around the location but this could easily be extended further with Wi-Fi, IR, sound, and video data. This work was part of a larger project to implement a patient telemonitoring solution.
One recent effort to combine sensor data is Google’s Project Tango [46
], announced 5 June 2014, and is an attempt to solve the issue of movement tracking using video, accelerometer and gyroscope data from a mobile device. Using such a device, researchers in Germany, again using Kalman filtering and Monte Carlo localisation combined with simple 2d floor plans, discovered that it is possible to track people indoors even in an environment that was not previously mapped, and to do so with accuracy [47
]. Although the authors do not offer insights into future enhancements, research such as this is not only applicable to tracking people through their environment but can be used by robotic assistants both for movement and tracking their owners. In the future, it is conceivable that this technology could be used to gather similar movement data. Once location data can be accurately determined there needs to be some way to detect activities themselves.
A high-level taxonomy of Activities of Daily Living (ADL) and their applications in healthcare and wellness in the home is proposed in Table 4
Activity detection is not only limited to the waking hours of a patient, in fact monitoring a patient’s sleep can be used to detect alert conditions as well as be an indicator for general wellness.
The NOCTURNAL system [48
] attempts to cater for the specific needs of individuals during the night-time period. Monitoring the sleep patterns of house occupants, specifically for restlessness, occupancy, and wandering, this system uses X10 technology for affordable yet flexible functionality. The system is comprised of pressure mats, lighting control, and actuators to assess and then intervene if abnormal sleep behaviour is detected; interventions include playing soothing music and dimming lighting. This information is fed back to a listening agent that can be used to monitor the occupant. In cases such as Obstructive Sleep Apnoea (OSA), sleep walking, and other sleep disorders, the real-time feedback given by such as system can alert carers to potential problems as they occur.
Another area of intense research is the delivery of healthcare remotely using technology as an enabler. Elderly people are the fastest growing segment of the population in developed countries but independent living comes with risks and challenges. In France alone it is predicted that the current 1:5 ratio of people over 60 years old will rise to 1:3 by 2050 bringing with it an increased demand on healthcare services and medical intervention. Telemonitoring and Telesupport are two ways to address this demand using technology. Solutions tend to use a multi-modal approach, employing several sensors to detect patient behavioural patterns and environmental conditions with the aim of providing intervention only when it is needed. Furthermore, solutions may be holistic, trying to encompass a wide range of monitoring and support functionality or conversely may be targeted specifically at one aspect of support, such as medication reminders, the actual implementation is often dictated by the patient’s condition.
First tested in lab conditions and followed up by application in the Broca Hospital in Paris, the system produced accurate data in 10 adults aged from 65 to 75 years old. Four parameters were used, sensitivity, specificity, error rate, and perfect classification when detecting abnormal situations with a 96% or above accuracy in sensitivity specificity and perfect classification, and only a 3% error rate. The 3% error rate could be reduced using these anomalous data to teach the system further. The system needs further trials on a larger population to improve accuracy and validate this approach but initial results are impressive.
Five main services are offered by the system: (1) caregiver Intervention i.e., to give the patient a required medication or respond to a non-emergency alarm; (2) cloud Storage to securely store medical data for data mining; (3) Emergency Response Services (ERS) in the case where immediate assistance is required; (4) real-time health advice and action to advise anyone present during an emergency situation the required response based on medical reports and past history; and (5) Patient Monitoring Service (PMS) to enable more independent living providing peace-of-mind to caregivers and relatives.
Although similar to [49
], the extension to include an extensible services framework and emergency response capabilities as well as caregiver and patient advice is unique. Implementing such a system will require a lot of further research into software and hardware solutions.