Advanced internet of things for personalised healthcare systems

TL;DR: This paper will give a systematic review on advanced IoT enabled PHS, and key enabling technologies, major IoT enabled applications and successful case studies in healthcare, and finally point out future research trends and challenges.

About: This article is published in Pervasive and Mobile Computing.The article was published on 2017-10-01 and is currently open access. It has received 301 citations till now. The article focuses on the topics: Wearable technology & The Internet.

Summary

I. INTRODUCTION

• Ecently, Internet of Things (IoT) is emerging as a new paradigm in information technology aimed at building up a dynamic global network infrastructure by connecting a variety of physical and virtual 'things' with the growing mobile and sensors.
• But these surveys most concentrate on examining individual layer of IoT enabled systems like sensing or data analysis, and lack of a systematic perspective review from the entire IoT eco-systems.
• Section II presents the background and current research of IoT enabled PHS.
• Section VI discusses research challenges and future trends.

II. CURRENT RESEARCH FOR IOT ENABLED PHS

• The initial vision of IoT was to extend the term "Internet" into the real world embracing everyday physical objects by means of Radio Frequency Identification (RFID) technology [2] [3] .
• So far, these sensory techniques are relatively technically and functionally sophisticated in manually controlled environments.
• The early work in mobile health focuses on developing specific algorithms for some diseases related data rather than general methods handling both health and medical data.
• In the IoT environment, PHSs are used by a large-scale population so that the scope of research in application layer has expanded into more wide areas, including healthcare service discovery, healthcare service composition, healthcare platform API, human-computer-interaction in healthcare, etc.
• As most of mobile devices are embedded a variety of inertial sensors (e.g., accelerometer, gyroscopes, etc.) and biomedical sensors (skin temperature, heart rate, etc.), they are designed for providing personalised and continuous cares for users.

A. Sensing and Identification Technologies

• Sensing and identification technologies target at recognizing physical objects and gathering human health information from sensors, tags, etc.
• Table 2 shows a list of wearable and ambient sensor categories.
• Applying both inertial sensors also enable accurately detecting a specific type of human motion and behaviors, such as bend knees, descend stairs [18] , ascend stairs or turning [57] .
• More importantly, these physiological devices are feasible to be used in out-of-hospital conditions, can enable a health data transmission through Internet.
• Image sensor in IoT enabled PHS usually indicate a camera that is utilised for recording and understanding human activities, emotions or other contexts by using image or video processing techniques.

Physiological sensors

• Systolic and diastolic blood pressure Electrocardiogram (ECG) [27].
• Rhythm and electrical activity of the heart Spirometer [76].
• Expiration, flow rate and lung volume Electrooculography (EOG) [77].
• Eye movement galvanic skin response (GSR) [15].
• Skin surface temperature Image sensors SenseCam [66].

Binary sensors

• Door open/close state Light switch [80] light on/off state Remote control switch [80].
• Remote control on/off state Location sensors Infra-red [81].
• Objects individual interact with NFC tags [84].
• Objects individual interact with that the attention of this paper is to review technologies for personalised healthcare system, the authors mainly summarize key wearable sensing technologies for IoT enabled PHS in this paper.

B. Networking Techniques

• Typically, networking layer in IoT applications contains a wide field of concepts and techniques, such as communication and location technologies, topologies, architecture, security and privacy, etc.
• Thus, the authors will mainly concern three research issues mentioned in Section II: topology, architecture, security and privacy.
• But for IoT enabled PHS, the topology needs to be a heterogeneous computing grids for collecting enormous amount of vital signs and health data, such as blood sugar, physical activity, blood pressure, oxygen saturation, etc. Viswanathan et.al [85] presents a new mobile grid computing topology 'hybrid static/mobile computing grid' for data-and patient-centric IoT enabled healthcare systems.
• Kart et al. [86] has applied SOA as a foundation to design, implement, deploy and manage health services in a distributed network system.
• In a high level view, data owners only need to specify access policies on the encrypted data; and their access control can be done automatically by the cloud.

C. Data Processing Techniques

• Data processing techniques for healthcare contains a quite wide scope regarding different types and format of data, different size of data, different purpose of applications.
• So here the authors only give a brief introduction of computational methodologies for health related data processing mentioned in section II, with a classification of data-driven approaches, knowledge base approaches and hybrid approaches, as shown in Table .

Category

• In IoT enabled healthcare field are based on a mechanism that makes use of a large volume of health related data from different subjects for training general models.
• Regarding the types of training or learning, it can be classified into supervised, semisupervised and unsupervised algorithms.

1) Supervised learning methods

• Dataset is often divided into training sets and testing set in the procedure of conducting supervised learning algorithms.
• Training dataset, also called labelled data samples is made use for building the prediction model, whilst a testing dataset is for validating the model.
• In recent years, deep learning has gradually became a popular method in medical diagnosis and health state classifications due to its high efficiency and accuracy.
• DT and ANN are combined in the study [25] for unconditional physical activity detections, and the results was prominently improved by being replaced every node in DT models with ANN.
• HMM was incorporated into Gaussian Discriminant Analysis (GDA) classifier presented in [124] , achieved a considerably satisfactory result to naïve Bayes and single GDA classifier in accuracy.

2) Semi-supervised learning methods

• Above supervised learning methods have their advantages on processing data in healthcare or clinical applications.
• So in these practical cases, semi-supervised or unsupervised learning methods are more popular.
• Furthermore, En-Co-training is an improved version proposed by Guan et al. [126] which is more flexible for physical activity measurements, since compared to Cotraining with two separately strong classifiers, En-Co-training trains data as a whole without requirement for confidence of the labelling of each classifier.
• A few studies investigated typical unsupervised clustering methods like K-means cluster [111] and Gaussian mixture model (GMM) [128] .
• As such, sensor-based activity data are regarded as stream of natural language terms to match objects for mining models from the web [131] .

2. Knowledge-based approaches

• Knowledge-based approaches represent and transfer knowledge from human expert (e.g., healthcare personnel and medical experts) into computer algorithms to establish computer-aided decision support system.
• Equally it can recommend diagnosis and clinical decisions to health personnel who will make changes in medication, and thus significantly improve the quality of live (QoL) for patients of chronic disease and elderly people living independently [1] .
• The knowledge model is expressed in some knowledge representation language or data structure that enable computer to execute the semantic rules.
• The conditions are specified in the antecedent, and the results of the reasoning are declared in the consequent.
• Also, knowledge-based decision support systems have been studied and deployed in various scenarios of remote health monitoring, reminding patients to visit physicians when their conditions are under severe situations.

3. Hybrid Approaches

• Since the training data samples and labels in nature environment are very difficult to obtain.
• The immature foundations often lead to erroneous predictions.
• While ontological methods are incompetent in handling a variety of uncertainties in real healthcare environment.
• COASR [118] is the typical case of combining two approaches in ADL detection for selfmanagement of elder people at their own homes.
• Collecting tens of thousands of training data for a large amount ADL classification is almost an impossible task in such environment, but based on mature techniques of classifying a few physical activities, with high level ontological reasoning, allows the issue well studied and application simply conducted.

IV. IOT APPLICATIONS AND CASE STUDIES IN PHS

• While the technologies applied in IoT enabled PHS are still in its early stage, the potential use in industry is rapidly evolving and growing.
• The authors review some successful platforms and applications including European projects, individually national projects and research approaches.
• The standalone device presented in the work was one of the most state-of-the-art techniques in early stage of activity monitoring investigation using wearable sensors.
• The platform version 2.0 resolves some practical issues such as storage, processor and battery life compared with version 1.0.
• The system trains data offline, but provides real-time feedback.

A. Physical activity platforms and applications MSP (Mobile sensing platform)

• WISDM (Wireless Sensor Data Mining) [137] is a typical platform that detects human physical activity based on Android phone sensors placed in one's pocket.
• Data is from the accelerometer, features are extracted according to the identification of time between signal peaks, and activities of walking, jogging, ascending stairs, descending stairs, sitting and standing are selected in this work due to their repetitive characteristics.
• MDurance [141] , a mobile healthcare support system for assessment of trunk endurance, is implemented in terms of the core functionalities of mHealthDroid.
• In healthcare, it studied gesture determination, including open/close hood, doors and trunk, checking steering wheel, etc. to assist doctors' diagnosis [123] .
• Etiobe [185] is another project devoted to treat child obesity.

B. Healthcare service with human interaction

• In recent years, self-management services in tele monitoring and AAL settings have been becoming a heated research and application facial point designed for satisfying user's specific requirements to improve the efficiency and success of a therapy (e.g., changing patient's dosage).
• Such systems are able to deal with a variety of patient conditions using sensor technologies, objective and subjective assessment methods, treatment plans and guidelines, with tailored information and advice being delivered to patients based on their feedbacks.
• Knowledgebased model is used as an inference agent describes objects and relationships in the sensing layer and hierarchically constructed.
• The approach was tested by a few elderly people and caregivers in Europe following the close loop principle.
• It uses simple approach that the elderly people can watch instructional videos of how to use modern household equipment through interacting with Near Field Communication (NFC) tags attached on the equipment with their smart devices.

C. CDSS automated prediction and diagnosis

• PredictAD (Predict Alzheimer's Disease) [148] is an European research project for developing a standardised and objective solution that would enable an earlier diagnosis of Alzheimer's disease, improved monitoring of treatment efficacy and enhanced cost-effectiveness of diagnostic protocols.
• The project develops a generic decision support software library and platform with different classification methods behind with a CDSS model composed by data tier for data collection and storage, logic tier for data processing and presentation tier for user interaction and interface.
• METEOR (Methodist environment for translational enhancement and outcomes research) [149] is an integrated clinical informatics framework that contains a data and logic storage EDW (enterprise data warehouse) and a clinical outcome prediction tool SIA (software intelligence and analytics) for physicians, caregivers and other clinical staff.
• The system is also designed to remotely monitor and control the patient's physical state from data collection of blood pressure, spirometry, pulse oximetry, temperature, etc. in the way of communications media like web browser and thus provide medical interventions and reminders.
• The whole engine integrates many key techniques like service-oriented architecture (SOA) and JBoss application server (JBoss AS) where manage reasoning rules extracted from electronic health record (EHR).

V. RESEARCH CHALLENGES AND FUTURE TRENDS

• Thus, so far, existing wearable sensing technologies are limited in terms of their size, fast response, continuous monitoring capability, wireless data transmission, and non-obstructive user experience.
• Any healthcare related applications must consider various security and privacy issues.
• Thus, it is highly important to effectively monitor and change individual behaviour with IoT enabled personalised healthcare systems, which requires a close collaboration between technical experts and clinicians.
• To succeed, these new models must extend their reach outside of the four walls of a clinician's office so that they can support patient behavior change beyond traditional clinician-patient interactions.

B. Future Research Trends

• Multiple sensors with different features often coexist in a single biometric system, also known as Sensing interoperability.
• Almost every biomedical sensor has its interoperability issues, few systems so far are able to handling with raw sensor data and feature extractions in pre-processing level in real, and thus expected to provided more practical and feasible approaches.
• The uncertainty of these data will be much higher than physical activity data training by classic machine learning methods in healthcare fields.
• IoT enabled PHS may be a specific application or service in the entire IoT environments, also known as Security and Privacy.
• But, the personalised health data will be stored and managed into the server of IoT systems.

VI. CONCLUSIONS

• Internet of Things paradigm represents the vision of the next wave of ICT revolution.
• IoT enabled technology in PHS will enable faster and safer preventive care, lower overall cost, improved patient-centered practice and enhanced sustainability.
• From a different perspective, the authors discussed current technology and infrastructure, such as sensing, networking and data processing technologies.
• More importantly, the authors provided a high level description of various IoT enabled healthcare applications.
• But, the authors are aware that the goals set up for IoT in healthcare are not easily reachable, and there are still many challenges to be faced and, consequently, this research field is getting more and more impetus.

Frequently asked questions

Q1. What are the contributions in this paper?

Copyright © and Moral Rights for the papers on this site are retained by the individual authors and/or other copyright owners. You may not engage in further distribution of the material or use it for any profit-making activities or any commercial gain. 

Q2. What are the future works in this paper?

To succeed, these new models must extend their reach outside of the four walls of a clinician ’ s office so that they can support patient behavior change beyond traditional clinician-patient interactions. Secondly, it is worthy to study of utilizing remote and self-care-oriented technologies to enhance the communication between patients and clinicians. B. Future Research Trends Sensing interoperability: multiple sensors with different features often coexist in a single biometric system. In this case, the future trend of using IoT technologies in PHS will focus on completely real life or namely uncontrolled environments. 

Q3. What are the main stages of knowledge-based methods?

Organizational knowledge model construction and rule-based inference are two main stages for carrying out knowledge-based methods. 

Q4. What is the key research issue of IoTtopology for PHS?

As the growth of connecteddevices and sub-networks, one key research issue of IoTtopology for PHS is how to transfer the heterogeneous static andmobile devices into hybrid computing grids. 

Q5. What are the benefits of IoT enabled technology in PHS?

IoT enabled technology in PHS will enable faster and safer preventive care, lower overall cost, improved patient-centered practice and enhanced sustainability. 

Q6. What are the main characteristics of knowledge-based approaches?

Among knowledge-based approaches, ontology is the most flexible and used approach in IoT enabled healthcare filed due to its reusability, computational completeness, decidability and practical reasoning algorithms. 

Q7. What is the role of the application layer in the IoT enabled PHS?

In the IoT environment, PHSs are used by a large-scale population sothat the scope of research in application layer has expanded intomore wide areas, including healthcare service discovery, healthcare service composition, healthcare platform API,human-computer-interaction in healthcare, etc. 

Q8. What are the advantages of semi-supervised learning methods?

2) Semi-supervised learning methodsAbove supervised learning methods have their advantages on processing data in healthcare or clinical applications. 

Q9. What is the idea candidate of future sensing technologies for IoT enabled PHS?

The idea candidate of future sensing technologies for IoT enabled PHS should be a tiny sensor into personal daily use items, including but not limited to clothing, watches, glasses, shoes, belts, and so on. 

Q10. What are the common types of wearable sensors?

The prominent development of low-cost and small-in-size wearable sensor such as inertial sensors (e.g., accelerator, gyroscope or barometric pressure sensors) and physiological sensors (e.g., spirometer, skin temperature sensor or blood pressure cuff), as well as wearable devices (e.g., fitness band or mobile phone) has facilities the process of measuring attributes related to individuals and their soundings. 

Q11. What is the key role of the application layer in the IoT enabled PHS?

But now in the IoT enabled PHS, the key role of specific application is mostly categorized into the application layer, thestudy focus of data processing layer here has transferred to generic algorithms to improve the accuracy and validity ofhealth data and or new data analytic tools to facilitate scalable,assessable and sustainable data structure. 

Q12. What are the common types of sensors used in medical devices?

As most ofmobile devices are embedded a variety of inertial sensors (e.g.,accelerometer, gyroscopes, etc.) and biomedical sensors (skin temperature, heart rate, etc.), they are designed for providingpersonalised and continuous cares for users. 

Q13. How did the experiments prove that the classifiers were accurate?

The experiments proved that using fusion of classifiers achieved high accuracy in the condition of extension of sensor network life time. 

Q14. What are some of the open resources of AAL systems that can share and reuse domain knowledge?

Some ontology based open resources of AAL systems that can share and reuse domain knowledge are already available such as SOUPA [114], SOPRANO [115], and GAIA [116]. 

Q15. What are the advantages of using unsupervised learning methods?

In addition to these, minority unsupervised learning methods have the aid of Intermediary to analyse abundant data resources from the web rather than directly labelling raw signals collected by the researchers. 

Q16. What is the main problem for IoT enabled healthcare applications?

for IoT enabled healthcare applications, a significant obstacle is that the majority of existing IoT enabled PHS system has limited permission on accessing and connecting hospital systems due to severe considerations on patients record and data. 

Q17. Why are there so many approaches and biomedical platforms proposed for sensing interoperability?

due to different types of sensors have diverse characteristics such as frequency, as such, many approaches and biomedical platforms have been proposed for sensing interoperability. 

Q18. What is the main idea behind the concept of a self-management service?

In recent years, self-management services in tele monitoring and AAL settings have been becoming a heated research and application facial point designed for satisfying user’s specificrequirements to improve the efficiency and success of a therapy (e.g., changing patient’s dosage). 

Q19. What are the adaptive methods for physical activity models?

Stikic et al. [109] made use of accelerometer and infra-red, compared different semi-supervised techniques, found that co-training and self-training methods are the most adaptive methods for physical activity models.