Assigment, A+ Work, No plagarism, On time, Original Work

wk5AIDEV.pdf

Home >Business & Finance homework help >Management homework help >Assigment, A+ Work, No plagarism, On time, Original Work

Computer Methods and Programs in Biomedicine 213 (2022) 106541

Contents lists available at ScienceDirect

Computer Methods and Programs in Biomedicine

journal homepage: www.elsevier.com/locate/cmpb

Application of artificial intelligence in wearable devices: Opportunities

and challenges

Darius Nahavandi a , Roohallah Alizadehsani a , Abbas Khosravi, Senior IEEE

a , ∗, U Rajendra Acharya, Senior IEEE

b , c , d

a Institute for Intelligent Systems Research and Innovation (IISRI), Deakin University, Waurn Ponds, VIC 3216, Australia b Department of Electronics and Computer Engineering, Ngee Ann Polytechnic, Singapore c Department of Biomedical Engineering, School of Science and Technology, Singapore University of Social Sciences, Singapore d Department of Bioinformatics and Medical Engineering, Asia University, Taiwan

a r t i c l e i n f o

Article history:

Received 5 October 2021

Revised 7 November 2021

Accepted 15 November 2021

Keywords:

Wearable devices

Healthcare

Machine learning

Deep learning

Internet of things

a b s t r a c t

Background and objectives: : Wearable technologies have added completely new and fast emerging tools

to the popular field of personal gadgets. Aside from being fashionable and equipped with advanced hard-

ware technologies such as communication modules and networking, wearable devices have the potential

to fuel artificial intelligence (AI) methods with a wide range of valuable data.

Methods: : Various AI techniques such as supervised, unsupervised, semi-supervised and reinforcement

learning (RL) have already been used to carry out various tasks. This paper reviews the recent applica-

tions of wearables that have leveraged AI to achieve their objectives.

Results: : Particular example applications of supervised and unsupervised learning for medical diagnosis

are reviewed. Moreover, examples combining the internet of things, wearables, and RL are reviewed. Ap-

plication examples of wearables will be also presented for specific domains such as medical, industrial,

and sport. Medical applications include fitness, movement disorder, mental health, etc. Industrial appli-

cations include employee performance improvement with the aid of wearables. Sport applications are all

about providing better user experience during workout sessions or professional gameplays.

Conclusion: : The most important challenges regarding design and development of wearable devices and

the computation burden of using AI methods are presented. Finally, future challenges and opportunities

for wearable devices are presented.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2. Paper collection strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

3. Brief review of wearables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

4. Denoising methods used in wearables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

5. Artificial intelligence methods. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

5.1. Feature extraction and engineering. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

5.2. Supervised learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

5.3. Deep learning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

5.4. Unsupervised learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

5.5. Semi-supervised learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

5.6. R einforcement learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

6. Application of machine learning algorithms in wearables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

6.1. Application of supervised learning methods in wearables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

6.2. Application of unsupervised learning methods in wearables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

∗ Corresponding author.

E-mail address: [email protected] (A. Khosravi).

https://doi.org/10.1016/j.cmpb.2021.106541

http://www.ScienceDirect.com

http://www.elsevier.com/locate/cmpb

http://crossmark.crossref.org/dialog/?doi=10.1016/j.cmpb.2021.106541&domain=pdf

mailto:[email protected]

https://doi.org/10.1016/j.cmpb.2021.106541

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

6.3. Application of semi-supervised learning methods in wearables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

6.4. Application of reinforcement learning (RL) methods in wearables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

7. Wearable applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

7.1. Sports. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

7.2. Healthcare. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

7.2.1. Fitness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

7.2.2. Health status monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

7.2.3. Helping with movement disorders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

7.2.4. Mental health . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

7.2.5. Autism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

7.2.6. Healthcare wearables shortcomings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

7.3. Industrial and manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

7.3.1. Examples of real-world industrial wearables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

7.3.2. Critical design factors for industrial wearables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

7.4. Human–robot interaction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

8. Wearable technologies challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

8.1. Data collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

8.2. Data transmission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

8.3. Security and privacy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

8.4. Localization quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

8.5. User adoption aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

8.6. Resource constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

8.7. Interoperability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

9. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

10. Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

Funding. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

Declaration of Competing Interest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

. Introduction

Wearables are small electronic and mobile devices, or com-

uters with wireless communication capabilities incorporated into

adgets, accessories, or clothes, which can be worn on the hu-

an body. There are also invasive versions of the wearables such

s micro-chips or smart tattoos [1] . Nowadays, different types of

earable devices have been invented. Some of the most common

earable devices are smart glasses and smart watches. The con-

umer market of the wearable devices is increasing steadily. The

earables are used to collect, transmit and even analyze the data

ommonly collected from the body of a human or animal. They

re purely mechanical devices or intelligent mechatronic systems

hich are commonly built using sensors, actuators, and compu-

ation parts. They can be used for early diagnosis and manage-

ent of medical conditions as well as measuring the vital signs

uch as body and skin temperature, blood pressure [2] , heart rate,

lectrocardiogram (ECG) [3] , and electroencephalogram (EEG) [4] .

ll these wearable devices are implemented with various technolo-

ies, capabilities and costs. People who use these technologies may

eed some skills to work with them.

Wearable devices are classified based on their requirement and

sage. Some of them are used according to the instructions of

hysicians to avoid serious problems. However, some wearables are

ot used in medical fields [5–7] . In [8] , a comprehensive review of

earable devices was done in which, smart wearable devices such

s watches, eyewear, headsets, jewelry, rings, chains, garments, and

racelets were described. The list of these devices can be seen in

ig. 1 [9] .

Wearable devices are made in different forms to meet their us-

ge requirements. They are commonly in small size while they are

xpected to sense continuously. They should be able to collect data

nd process them to improve the quality of life. Therefore, wear-

bles need to communicate in a secure way while keeping their

ower consumption as low as possible. The security of wearable

evices is a big challenge. They may be able to collect the data lo-

ally or send them to an external device. In both cases, the data

hould be encrypted to enforce their privacy. Given that wearable

evices usually have low computational power, a lightweight au-

hentication test is needed. In addition, wearable devices must be

ble to communicate in real-time; such requirement impacts on

he challenge of power consumption management.

The motivation behind this review is the fact that the emerg-

ng field of wearable devices has the potential to open new ap-

lication opportunities in various domains. The focus of this re-

iew is on medical, industrial, and sport applications. We focused

n the medical domain since it is directly related to the lives of

eople. With enough development, wearables have the potential to

evolutionize the medical domain leading to cost-effective health-

are services and longer lifetimes. In the industrial domain, wear-

ble devices can make workstations more ergonomic. To this end,

he workers can be equipped with appropriate wearable devices

n order to accelerate the industrial processes leading to short-

ned working hours and better psychological health. The sport do-

ain is also important since it can be used for medical diagnosis

nd treatment. Additionally, sport is directly related to the general

ell-being of society. Therefore, sport domain is also an important

omain that is worth reviewing.

A comprehensive review of existing wearables, their capabilities

nd shortcomings can shape future research directions. In this re-

iew, wearable devices as well as the role of AI methods to achieve

arious tasks with wearables are investigated. The employed pa-

er collection strategy is outlined in Section 2 . Various wearables

re reviewed briefly in Section 3 . Denoising methods used in wear-

bles are explained in Section 4 . Feature extraction, engineering

nd artificial intelligence methods used are described in Section 5 .

he applications of machine learning in wearable devices are pre-

ented in Section 6 . Applications of wearables are explained in

ection 7 . The challenges of developing wearables are reported in

ection 7.4 . Discussion and future insights will be presented in

ection 9 and the conclusion is given in Section 10.

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

Fig. 1. Different categories of wearables used.

. Paper collection strategy

For reviewing previous papers in the field of wearables, all

atasets supported by Google scholar such as IEEE, Science Di-

ect, Springer, ACM Digital Library, and Hindawi were searched. The

earch query used in this work was:

(wearable technology healthcare OR wearable devices OR wear-

bles OR smart wears OR wearable technology OR industrial wear-

ble OR sports wearable) AND (artificial intelligence OR data min-

ng OR deep learning OR machine learning).

Three authors inspected the papers collected based on the

earch phrases mentioned above. Papers that two out of three au-

hors agreed on their relevance to this review were selected for

urther analysis. Based on the above phrases and the opinion of

he three authors, 132 papers published in high-ranked journals

nd conferences were selected. The paper selection mechanism is

resented in Fig. 2 in which the number of selected papers from

ach publisher is reported separately.

We did not limit our search to medical applications since

earable devices used in different domains share common hard-

are/software technologies. Hence, it was necessary to take a

roader view during data collection for this review. In addition to

edical domain, this review considered industrial and sport ap-

lications of wearables as well. The motivation was that industrial

nd sport applications are closely related to medical applications.

n industrial domain, well designed wearables can be used to make

he working environment more ergonomic reducing workers’ in-

uries. In sport domain, wearable devices can help individuals with

heir fitness programs leading to healthier society.

The selected papers were carefully studied by the three authors.

he study of papers related to each of three major fields (medi-

al, industrial, and sport) was assigned to one of the authors. Each

uthor extracted and organized the necessary information for this

eview. The investigation of the selected papers was primarily fo-

used on the main approach used during software/hardware im-

lementation of the wearable devices.

Certainty about the outcome of the reviewed papers is verifi-

ble since most of these papers have been published in top-ranked

ournals and conferences. Additionally, many of the reviewed pa-

ers or wearable devices have already been used in practice prov-

ng their capabilities.

. Brief review of wearables

Wearables can be used for data collection in daily activities,

port performance, and health monitoring. There are different

ypes of wearables such as smartwatches, hearing aids, electronic

attoos, wristbands, subcutaneous sensors, head-mounted displays,

lectronic textiles, and footwear as shown in Fig. 3 (a) [10] . These

evices are placed on different body parts to measure electrophys-

ological and biochemical signals or deliver drugs.

Wearable devices are used for augmented, virtual, and mixed

eality, artificial intelligence, and pattern recognition [ 11 , 12 ]. These

echnologies commonly contain microprocessors and sensors. Ad-

itionally, these devices are usually capable of recording data and

xchanging them over wireless connections [13] . Sensors used

n wearable devices include barometers and inertial measure-

ent unit (IMU) which is combination of gyroscopes, accelerom-

ters, and sometimes magnetometers. Optical sensors needed in

pectrophotometers, cameras, chemical probes, electrodes, micro-

hones, shock detectors, and pressure sensors are other types of

ensors used in wearables [14] . Utilizing the sensors in multiple

earables provides a rich collection of data which can be anal-

sed by researchers or used by experts to provide medical treat-

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

Fig. 2. Paper selection mechanism.

Fig. 3. Wearables devices used to monitor physiological parameters: (a) different wearable devices which have been designed for different parts of human body, (b) various

technologies used to transfer data collected from wearables to other devices.

“

ent remotely. These data can be transmitted by different types

f networks [15] . As shown in Fig. 3. b, these networks can even

ransmit the data over the internet. The capability of wearables to

perate in a network of connected devices paves the road towards

mplementing Internet of things (IoT) [16] .

The existing wearable devices can be categorized based on their

pplications and the body parts on which they are mounted. Since

016, the distribution of number of wearables in different appli-

ation domains is shown in Fig. 4. a [17] . As can be seen, the

lifestyle” application has the highest number of wearables (about

00) while the “pet animals” application has the least number of

xisting wearables. Since 2016, the distribution of existing wear-

bles based on their target body part is shown in Fig. 4. b [17] .

t is clear that most of the existing wearables are mounted on

he head (about 65 devices), followed by torso and neck and then

ody parts have the third highest number of wearables. Fig. 5

llustrates two samples of wearables with medical applications.

ig. 5. a shows a typical head-mounted wearable used for EEG anal-

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

Fig. 4. Distribution of number of existing wearables (as of 2016) based on: (a) application domains, (b) target body parts.

Fig. 5. Illustration of two typical wearables with medical applications: (a) head-mounted wearable device for EEG measurement, and (b) torso mounted wearable device for

ECG measurement.

sis and Fig. 5. b shows a torso-mounted wearable used for ECG

nalysis.

. Denoising methods used in wearables

Signals, which have been gathered from wearable sensors, are

ommonly affected by noise. The noise sources are generated when

he measuring element and the data collection system try to col-

ect the signals. This section aims to introduce the AI-based hard-

are designed for denoising. As real information generated by bi-

logical systems, randomness of these systems are relatively low

hile the real information collected in time are often correlated.

For denoising, researchers have used different methods. In [18] ,

sixth-order bandpass IIR filter was used to eliminate the noise.

n another research [19] , a deep convolutional neural network was

sed. In [20] , the noise and baseline removal of all ECG signals

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

Fig. 6. Different AI methods and their main subfields.

[

(

ere performed with Daubechies wavelet filters. In [21] , as one

f the pre-processing steps, an 8-point moving average filter was

sed to remove noise. To this end, finite windows of moving aver-

ge filter were convolved with the signal. It took an average of the

utput signal for discrete-time noise reduction and enhanced the

eak value identification.

Chen et al. [22] developed a statistical model to simulate struc-

ured noise processing derived from a wearable sensor. Synthetic

ata generated using a structured noise model was studied and a

actor analysis-based method was proposed for denoising. Lee et al.

23] proposed a denoising method that references photoplethys-

ography to alleviate intrinsic and extrinsic noise in electrodermal

ctivity. Their method attenuates the extrinsic noises by applying

everal filters such as high-pass and wavelet filters. Then, intrinsic

espiration noises were detected and attenuated by a subject inde-

endent machine learning model that could detect noise.

A new electrocardiogram (ECG) denoising technique was pro-

osed in [24] . In their work, denoising was done by variable fre-

uency complex demodulation algorithm. To remove the noise,

his algorithm is used to perform the sub-band decomposition of

he noise-contaminated ECG. More improvement in ECG quality is

one by not only removing baseline drift but also smoothing via

daptive mean filtering. Two datasets were used to validate the

roposed method. The performance of the proposed denoising al-

orithm was compared with other denoising algorithms and its su-

eriority with respect to other methods is shown.

. Artificial intelligence methods

In this section, at first, feature extraction and engineering is re-

iewed and then different categories of AI methods i.e. supervised

earning, unsupervised learning, semi-supervised learning and re-

nforcement learning are introduced. These learning methods and

heir main subfields are shown in Fig. 6 .

.1. Feature extraction and engineering

Feature extraction is one of the fundamental steps in machine

earning. Having too many features could easily confuse the ma-

hine learning algorithms [ 25 , 26 ]. Therefore, feature selection al-

orithms are used to select the clinically significant features. The

ean and mode [27] or algorithms such as principal components

nalysis (PCA) [28] , linear discriminant analysis (LDA) [29] , inde-

endent component analysis (ICA) [30] , locally linear embedding

LLE) [31] , and autoencoders [32] can also be used to select the

tatistically significant features. Such features can be exploited by

earning methods.

The process of extracting useful features from raw data based

n domain knowledge is called feature engineering [33] . The first

tep in feature engineering is developing useful features by (i) au-

omatic, (ii) manual, or (iii) fusion of both manual and automated

eature extraction. The next step is feature selection in which a

ubset of extracted features is selected according to some feature

coring measure. The performance of selected features is then eval-

ated based on the target dataset. This process is repeated until

atisfactory results are obtained.

.2. Supervised learning

The learning algorithms are divided into two main types: super-

ised and unsupervised. In supervised learning, the desired output

or the training samples is known and the model is trained using

he given samples of data and their desired outputs [34] . Generally,

upervised learning is used for classification in which the goal is

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

[

(

[

o map an input sample to the output label [35] . It is also used for

egression whose goal is learning a mapping from inputs to con-

inuous output. In both classification and regression, we want to

nd the correct relationships between the input and output. In-

eed, we are looking for a model that can produce correct out-

ut data effectively. If the training data are noisy or have incorrect

abels, the effectiveness of the trained model will be clearly de-

raded. Some of the common supervised learning algorithms are

upport vector machine (SVM) [36] , artificial neural network (ANN)

37] , Naïve Bayes [ 38 , 39 ], and random forest [40] .

.3. Deep learning

Deep learning (DL) is part of a broader family of machine learn-

ng methods based on artificial neural networks (NNs). In the

ealm of deep learning, we often come across convolutional neu-

al network (CNN) which is a special type of NN capable of han-

ling 2D image data [ 41 , 42 ]. The primary component of a CNN is

he convolutional layer which performs convolution on a given im-

ge. To this end, one needs to specify a 2D array of weight values

alled a kernel which is smaller than the image. The convolution

peration is simply the dot product of the kernel with a kernel

ized patch of the given image [43] . The convolutional layer out-

ut is passed through an activation function such as ReLU. 1 Auto-

atic feature extraction is one of the most important characteris-

ics of CNNs. However, training CNNs usually demands high com-

utational resources. In recent years, such a burden has been al-

eviated due to the advent of powerful graphics processing units

GPUs) [44] .

.4. Unsupervised learning

In unsupervised learning, the objective is to learn the inherent

tructure of unlabelled data. The most usual tasks within unsuper-

ised learning are clustering, density estimation, and representa-

ion learning. For this purpose, some of the algorithms such as

rincipal component analysis (PCA) and auto-encoders have been

roposed [45] . Exploratory analysis and dimensionality reduction

re two common use cases used in unsupervised learning. In sce-

arios where the dataset analysis is impossible for humans; un-

upervised methods can be used to gain initial insights into the

ata. The insights can be used for testing different hypotheses. For

imension reduction, the data are represented by fewer features.

his process can also be done using unsupervised learning. To this

nd, the relationship between features must be discovered. It can

elp us to eliminate the redundant features. Consequently, process-

ng the data can be done by a much less intensive solution [46] .

.5. Semi-supervised learning

In scenarios where the number of labelled samples is small

hile number of unlabelled samples is large, supervised and un-

upervised learning cannot be used effectively. In this situation,

emi-supervised learning algorithms can help. They can be trained

y a small number of labelled and a large number of unlabelled

ata to predict a new example. When there are some labelled data,

hey can help the algorithms to use the unlabelled data more ef-

ciently and produce considerable improvement in learning accu-

acy. Acquisition of labelled data to be used in learning problems

ommonly requires expert agents. Labelling the samples may be

ostly and in some cases impossible due to large number of un-

abelled samples. Under these circumstances, the importance of

emi-supervised learning becomes clear [47] .

1 Rectified Linear Unit

.6. R einforcement learning

The reinforcement learning (RL) is learning to map situations to

uitable actions such that a numerical reward signal is maximized

48] . Unlike supervised learning, in RL, the learner is not provided

ith the desired action and it has to try different actions in differ-

nt situations (also known as states) to figure out the best actions

eading to the maximum reward given the observed states. It is

mportant to learn action selection such that the long term utility

s maximized since naively choosing to maximize the immediate

eward might lead to suboptimal performance in the long run. RL

roblems can be modelled as Markov decision processes (MDPs). A

DP is a 4-tuple ( S , A, P, R ), where:

• S is the set of states (state space) • A is the set of actions (action space) • P ( s t+1 = s ′ | s t = s, a t = a ) is called the transition function which

outputs the probability of observing state s ′ at time step t + 1

provided that at time step t observed state is s and chosen ac-

tion is a . • r t + 1 = R ( s t = s, a t = a, s t + 1 = s ′ ) is the expected reward if at

time step t, the observed state is s and execution of the chosen

action a will lead to state s ′ in the time step t + 1.

. Application of machine learning algorithms in wearables

Various machine learning methods have been used in the field

f wearables. In this section, some of existing works which com-

ine wearable devices with machine learning algorithms are re-

iewed. The review has been categorized based on type of machine

earning methods. The summary of works done using wearables

nd AI techniques are shown in Table 1 .

.1. Application of supervised learning methods in wearables

Supervised learning methods are widely used in machine learn-

ng to develop the automated systems. Saadatnejad et al. [49] sug-

ested a novel electrocardiogram (ECG) classification algorithm. On

earable devices, this method was used for continuous monitor-

ng of cardiac disease. The advantage of this method was its low

ower consumption. Their method used multiple long short term

emory (LSTM) recurrent neural networks and wavelet transform.

heir method achieved high ECG classification performance. Simi-

arly in [50] , a novel ECG classification algorithm was proposed and

sed in low-power wearables based on spiking neural networks.

spike-timing dependent plasticity and reward-modulation were

mployed in which the model weights are trained according to the

imings of spike signals. The results showed that it was suitable for

eal-time operation. Additionally, in the real-time classification of

CG signals, its energy consumption was significantly lower than

ther similar devices. In another work reported by Acharya and

asu [51] , the primary objective was to build classification mod-

ls to identify anomalies of patients’ breathing sounds. These data

ere used for automated diagnosis of respiratory and pulmonary

iseases. A deep learning model was used to classify respiratory

ounds. Additionally, a local log quantization strategy was pro-

osed to reduce the memory footprint which can be used in mem-

ry constrained wearable devices.

Wearable sensors can be used in disease diagnosis based

n physical movement of patients. For instance, Hssayeni et al.

52] used a LSTM recurrent neural network (RNN) to detect early

igns of Parkinson’s disease (PD) using accelerometers and gyro-

copes data. In another study, waist-worn accelerators and SVM

ere used to detect freezing of gate (FoG) experienced by PD pa-

ients [53] . The walking pattern can be used to diagnose Alzheimer

isease. Varatharajan et al. [54] monitored walking patterns of pa-

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

Table 1

Summary of works done using wearables and AI techniques.

Learning category

Reference # Algorithm Name Application Supervised Unsupervised Semi-supervised Reinforcement

[68] SVM Construction Workers’ Stress Recognition √

[69] ANN Heartbeat Classification √

[70] Statistical analysis Activity recognition √

[71] Decision Trees Pharmacotherapy Management for

Parkinson’s Disease Patients

√

[72] Long Short Term Memory

(LSTM)

Activity Recognition √

[73] Random Forest Physical Fatigue Detection √

[74] K-means Telemonitoring of patients with

Parkinson’s disease

√

[75] K-Means Human Activity Recognition √

[76] Spectral Clustering,

hierarchical clustering

Human activity recognition √

[77] K-Means Human activity recognition √

[78] Expectation Maximization

(EM) algorithm

Activity recognition √

[79] K-means Detection of Poor Posture √

[80] Simple 1-NN classifier and

SVM

Location recognition from wearable video √

[58] LSTM Cardiovascular risk prediction √

[81] SVM Human activity recognition √

[62] Random forest Activity recognition √

[63] Genetic algorithm Categorization of brain signals √

[82] Convolutional neural

networks

Human activity recognition √

[83] Deep reinforcement

learning

Musculoskeletal modeling and locomotion

analysis

√

[84] Reinforcement Learning Activity recognition √

[85] Inverse Reinforcement

Learning

Activity forecasting √

[86] Deep Reinforcement

Learning

Individualized Treatment Planning for

Parkinson’s Disease

√

[87] ANN Activity recognition √

[88] Inverse Reinforcement

Learning

Activity forecasting √

ients using a dynamic time warping algorithm and several wear-

ble sensors including accelerometers. Based on the perceived

alking pattern, early signs of Alzheimer disease were detected.

.2. Application of unsupervised learning methods in wearables

In [55] , Das et al. proposed an unsupervised learning approach

or heart-rate estimation from electrocardiogram (ECG) data col-

ected by wearable devices. Spatio-temporal properties of ECG sig-

als were encoded directly into spike training. In the next step,

he spike training was used to excite recurrently connected spik-

ng neurons in a liquid state machine computation model. An un-

upervised readout based on fuzzy c-Means clustering of spike

esponses was designed using particle swarm optimization. Their

roposed method was easily implemented on spiking-based sys-

ems. The method advantages are its high accuracy and signifi-

antly low energy footprint. Consequently, the battery life of wear-

ble devices was extended. Another unsupervised learning algo-

ithm was proposed by Krause et al. [56] . In this work, without

xternal supervision, an online wearable system was designed, im-

lemented and evaluated. It could determine the context of typical

ser and probabilities of context transition. They used statistical

nalysis and machine learning in their graph algorithm techniques.

he results showed that their proposed method could determine a

ser context model while it only required data from a device with

hysiological sensors.

In [57] , a new version of unsupervised deep learning was pro-

osed which optimized the data during preprocessing in wearable

ensors. It only needed 11.25 ns as its computation time. Its recog-

ition performance has been improved for feature selection and

xtraction. A new technique for data analysis has been introduced

o minimize the computation time.

.3. Application of semi-supervised learning methods in wearables

Wearable devices have the potential to collect huge amounts of

ata. However, labeling these data is costly and time-consuming.

herefore, it is desirable to devise methods to exploit unlabeled

ata while reducing labeling costs as much as possible. Semi-

upervised approaches are promising approaches to use a mix of

imited labeled data and a large volume of unlabeled data ef-

ciently. Ballinger et al. [58] used off-the-shelf wearable heart

ate sensors to collect data from numerous participants across the

orld using a mobile phone application. The objective was to de-

ect multiple medical conditions such as diabetes, high cholesterol,

tc. using a multi-task LSTM. Two semi-supervised approaches

ere proposed that could outperform hand-engineered biomark-

rs from the medical literature. In the first approach, a LSTM was

re-trained as a sequence autoencoder. The pre-trained parame-

ers were used to initialize a second supervised phase using pool

f limited labeled data. In the second approach, the synthesized

ataset was used for pre-training.

In [59] , a novel method to automatically detect near-miss falls

ccording to a worker’s kinematic data was proposed. These data

ere captured from wearable inertial measurement units (WIMUs).

semi-supervised learning algorithm was proposed to learn from

he data. This algorithm was a support vector machine (SVM)

hich was designed for near-miss fall detection. For collecting the

ear-miss falls, two experiments were conducted. These data were

sed to test the proposed approach. This WIMU-based method can

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

e used to identify ironworker near-miss falls without disrupting

obsite work and can help to prevent fall accidents.

Stikic et al. [60] introduced a new method for activity recog-

ition. In a semi-supervised learning process, small amounts of

abeled data were combined with unlabeled data. Their proposed

ethod propagated information in a graph that contains both la-

eled and unlabeled data. Based on feature similarity, two differ-

nt ways were introduced to combine several graphs. The quality

f the label propagation process and the performance of classifiers

ere evaluated in their research.

For activity recognition, in [61] , the feasibility of semi-

upervised learning was tested to reduce the level of supervision.

wo semi-supervised techniques (self-training and co-training)

ere used to learn activity models from few labeled data. The

esults of this work demonstrated that co-training worked better

han self-training because it used additional information from sen-

or modalities during the training process. In addition, in some

ases, it even could achieve better recognition accuracy compared

o the fully supervised approaches. Their proposed method was

onducted based on a pool-based setting. Accordingly, a large num-

er of unlabeled training data were available in addition to a small

et of labeled training data. The algorithm was able to select the

est informative samples which were then labeled by an expert.

nother work on human activity recognition is LabelForest [62] .

ata collected from humans via wearables are often accompanied

y a significant amount of noise and uncertainty. LabelForest is

non-parametric semi-supervised learning framework for activity

ecognition which improves the performance of ML algorithms by

xpanding the training set. LabelForest chooses a subset of unla-

eled data for labeling. The sample selection is done based on sim-

larity with the labeled samples. LabelForest framework is made

f two algorithms: 1. spanning forest algorithm for sample selec-

ion and labeling, and 2. silhouette-based filtering method to select

amples with more confident clustering assignment for inclusion in

he training set.

Wiechert et al. [63] collected EEG brain signals using a wear-

ble headband called Muse from participants performing different

asks such as reading, listening to music, etc. The objective was to

dentify participants and the type of activity they were perform-

ng when EEG signals were being recorded. To this end, K-medoids

ith an evolutionary algorithm were combined to perform multi-

bjective clustering. The genetic algorithm was used to find the

ost appropriate K medoids. Wiechert et al. reported that their

ethod could outperform K-means.

.4. Application of reinforcement learning (RL) methods in wearables

RL has also found its way into the field of wearable technolo-

ies. ADAS-RL [64] is a modified version of Q-learning algorithm.

t not only integrates the behaviors but also the reactions of the

river to adapt and tune the warning interventions of Lane Depar-

ure Warning System (LDW) continuously. The proposed method

s able to track any changes in driving behavior and adapt the

requency of warnings allowing drivers to stay within a reason-

ble distance i.e. about 1.75 m from lane markings. FaiR-IoT (In-

ernet of Things) [65] is another RL-based framework using Q-

earning for adaptive and fairness-aware human-in-the-loop IoT

pplications. The method was evaluated on a human-in-the-Loop

mart house IoT application and human-in-the-Loop automotive

dvanced driver assistance system. In the smart house application,

he objective was to control the home thermostats automatically

y monitoring human body temperature changes over time. The

ob of the driver assistance system is to alert the driver when there

s a risk of colliding with an object in front of the vehicle. Stan-

ard forward collision warning systems measure the time-to-crash

ased on the distance and the relative velocity of the object in

ront of the vehicle. If the time-to-crash is below a certain thresh-

ld, the system alerts the driver to apply the brakes. However, a

etter approach is making the threshold adaptive based on the

river characteristics such as his/her response time and whether

e/she is distracted or not. FaiR-IoT has taken such factors into ac-

ount to adjust the time-to-crash threshold dynamically.

In the field of medical applications, some patients are in need

f constant monitoring. Wearable sensors can make the monitor-

ng process easier without the need for keeping the patients in

he hospital for long periods. Baucum et al. [66] have proposed

data-driven RL framework to optimize PD medication regimens

ased on wearable sensors data. They conducted their study us-

ng a dataset of 26 PD patients who wore wrist-mounted move-

ent trackers for two separate six-day periods. The patients’ med-

cation regimens were modified based on physician evaluations of

ollected data after the first wear period. The method was imple-

ented in two steps:

1. A simulation model was built and evaluated. It provides in-

formation about individual patient’s movement symptoms re-

sponse to medication administration.

2. The simulation model was used for training an RL agent using

policy gradient [67] . The trained agent was able to recommend

optimal medication types, timing, and dosages during the day,

while incorporating human-in-the-loop considerations on med-

ication administration.

The existing literature on wearable technologies is not limited

o the above paragraphs. The summary of works done using wear-

bles and AI techniques is shown in Table 1 .

. Wearable applications

Wearables have already emerged in various application do-

ains such as eyewear, sport trackers, healthcare, industry, etc.

he wearable technologies are not exclusive to medical applica-

ions and they share common hardware/software across different

omains. Therefore, reviewing wearables in different domains pro-

ides a broader perspective about wearable technologies. To gain

etter insight on the potential applications of wearables, applica-

ions related to healthcare, manufacturing, and industrial domains

re reviewed. Some of existing wearable devices are presented in

able 2 . Although performance statistics of wearable devices have

ot been the primary concern of this review, some performance

tatistics ( Fig. 4 , Fig. 7 , and Fig. 8 ) have been restated here from

he reviewed papers.

.1. Sports

In the sport applications, the wearable help the players to im-

rove their skills in their favorite sports. Existing wearables in

port applications is shown in Table 2 . Injury prevention is criti-

al in any sport and wearable devices may be used to avoid po-

ential injuries while players are enjoying their favorite sports.

or example, Chen et al. [89] developed a fuzzy logic inference

ystem which receives data such as temperature, humidity, etc.

rom wearable devices and determines the wearer’s heat stroke

ossibility. Their approach can detect the possibility of suffering

rom heat stroke and the wearer can be alerted in time. Skazal-

ki et al. [90] used commercially available wearable devices to

onitor functional movements, heart rate, and workloads of vol-

eyball players. The collected data can be used to maximize the

layers’ performance and at the same time minimize possible

njuries.

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

Table 2

Examples of existing wearables in various application domains.

Application Wearable device name Device description

Eyewear Chromatic smart glasses It is a smartwatch which has built-in activity tracker for fitness purposes, wireless charging

capability, and HD camera for taking point of view (POV) photos.

Vufine + HD Wearable Display Vufine + is a high definition wearable display which can be connected via HDMI to smartphone,

laptop, or drone. Vufine + can be used as a second monitor and it is clear enough for video and text

display.

Omni-Wearable Action Cam

Sunglasses

This device is sunglasses which features an integrated HD video camera to record the events

happening around the user.

Focals by North This device is smart glasses which offer augmented reality (AR). The goal of the device is displaying

the notifications from the user’s phone directly into his/her field of view.

Training

assistants

GAME GOLF Digital Shot Tracking

System

Accurate GPS shot tracker which is designed to help Golf players improve their shots and playing

experience.

Basketball Replay Analyser by

Blast Motion

This device records the actions of Basketball players and provides them with performance metrics

so that the players can improve their performance. The device sensor which is attached to the waist

of the body has wireless connection to an IOS device.

Marlin Marlin is a waterproof device that helps swimmers with their training, open water navigation, and

tracking.

T-Goal Wearable Soccer Data

Tracker

A compact device that can track speed, distance, sprints, and positioning of soccer players as they

play.

Phoenix A medical exoskeleton which helps people with mobility disorders. The device makes it possible to

stand up and walk. The device has two actuators at its hip. The knee joints provide support during

stance and ground clearance during swing.

Alex posture system This device monitors the posture of the user neck in order to improve his/her neck posture.

Relaxation Beddr SleepTuner The sleep tuner helps the user to find out the main causes of poor sleep. The device can measure

blood oxygen levels, heart rate, amount of time in bed, etc.

BrainLink Lite V2.0 A headset which helps the user with focusing and training, meditation and pressure relief.

Vigo The Stimulating Headset A Bluetooth headset that monitors the driver’s eyelid motion to assess his/her level of drowsiness.

Lumos Smart Sleep Mask A device that helps the travellers fight jet lag. To this end, the device transmits short light pulses to

adjust the body clock.

Smartwatches Aipower Wearbuds Basically, this device is the integration of headphones into smartwatch.

Garmin Approach S20 Golf Watch A device for training golf which features a high-sensitivity GPS and provides the player with useful

distance data. Using these data, the user can improve his/her shots. The device also provides daily

activity tracking and smart notifications (e.g. incoming calls and messages).

Parkinson Smartwatch A smartwatch that tracks Parkinson’s disease. The patient records his/her condition changes using

this device. The recorded data during the day is stored in a cloud service which is accessible

anywhere in the world by the patient and his/her doctor. Based on the recorded data, the doctor

prescribes optimal dosage of the medicine for the patient.

Smartbands Misfit Ray The device automatically tracks the fitness and sleep metrics of the user. For example, number of

taken steps, travelled distance, burned calories, and light/heavy sleep are tracked.

Samsung Galaxy Fit The device provides fitness statistics such as heart rate. The device also allows the user to reply

instantly with pre-set messages for incoming texts.

Xiaomi Mi Smart Band 4 Xiaomi Mi Smart Band 4 provides health monitoring such as heart rate. It has multiple tracking

modes such as Treadmill, outdoor running, cycling, etc. which can be used during sport activities.

Incoming calls, messages, and calendar notifications are also supported.

Huawei Band 3 Pro Huawei Band 3 Pro-supports notifications for incoming calls and messages as well as playing music.

It also offers sport-related functionalities such as step count, calories burned, etc.

E-patches Lief A wearable device that is designed for stress relief. Using Lief, the users learn how to train their

body to stay calm and focused. Lief is an ECG smart patch which improves heart rate variability.

Heart rate is a scientifically-proven biomarker of stress.

Mesana This is a sensor patch which addresses issues within circulatory diagnostics and

cardiovascular-prevention.

VivaLNK Vital Scout VivaLNK Vital Scout is wearable patch that measures stress and recovery rates using medical-grade

ECG sensors.

Wearable Ultrasound Patch This is a patch which can measure internal blood pressure e.g. blood pressure inside deep arteries,

heart, or lungs.

.2. Healthcare

Considering that wearable devices are worn by their users,

hese devices have a lot of potential in providing mobile health-

are services. Different types of wearables e.g. smartwatches, on-

ody cameras [91] , masks, E-patches, etc. have been developed for

ealthcare applications. The most common type of data measured

y wearable for healthcare purposes include heart rate, blood

ressure, body temperature, blood oxygen saturation, posture, and

hysical activities.

.2.1. Fitness

Some of these devices are designed to track fitness-related

ctivities. For example, chromatic smart glasses are an eyewear

evice which tracks the user activity for fitness purposes. Xi-

omi Mi Smart Band 4, Misfit Ray, and Samsung Galaxy Fit are

martbands which provide fitness statistics. A short description

f these devices is available in Table 2 . Fitness-related wear-

bles can motivate their users to increase their activity and

ecome healthier but their measurements might not be accu-

ate always. For example, Dooley et al. [92] have evaluated the

erformance of three wearables called Fitbit Charge HR, Apple

atch, and Garmin Forerunner 225. The experiment was con-

ucted using 62 participants with age between 18 and 38 years

ld. The three devices were used to measure the heart rate

nd energy expenditures of the participants. They reported that

he accuracy of these three devices were within the acceptable

ange.

.2.2. Health status monitoring

Some other wearables can be used to monitor the health sta-

us of their users. These devices may be capable of forecasting the

otential health issues of people wearing them even before they

eel sick or discomfort due to those issues. These devices may also

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

Fig. 7. Distribution of 25 enterprises surveyed in [116] for each of the four aspects: (a) industrial sector, (b) application scenarios, (c) data processing methods, and (d) data

interaction level.

ake a step further and inform the doctor automatically [93] . It is

eedless to say that in some cases, early diagnosis of diseases can

e lifesaving. Type of healthcare wearables such as Mesana is a

ensor patch which addresses issues within circulatory diagnostics

nd cardiovascular prevention. Another example is a wearable ul-

rasound patch with the ability to monitor blood pressure in deep

rteries. Two ECG-based patches are Lief and VivaLNK Vital Scout

an be used. Lief helps to improve the heart rate variability which

s useful for stress relief. VivaLNK Vital Scout helps to measure

he stress and recovery rates. Two more ECG-based wearables have

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

Fig. 8. Result of evaluating three world-leading groups: (a) wearable computing, (b) wearable computer systems, and (c) contextual computing based on five factors (er-

gonomic product design, data interaction on device, operational stability, external software integration, external hardware integration).

(

een developed to monitor the patients with heart disorders by

inokur et al. [94] and Yang et al. [95] .

Controlling the condition of patients who suffer from chronic

iseases is as important as early diagnosis of diseases. For ex-

mple, Parkinson patients may need to receive variable dosage of

heir medicines based on their body condition. The prescription

f medicine dosage must be done by medical experts. However,

ccess to experts may not be possible all the time. Hence, wear-

ng Parkinson smartwatch helps to record the patient’s noticeable

hanges in his/her condition throughout the day. The recorded data

an then be sent to a medical expert to receive appropriate dosage

f the required medicines. Another Parkinson-related wearable has

een developed by Lin et al. [96] helps to assess bradykinesia

everity of Parkinson patients based on ten-second whole-hand-

rasp action. Delrobaei et al. [97] proposed an objective dyskinesia

core based on motion capture data obtained from a mobile full-

ody wearable system equipped with inertial measurement unit

IMU).

Patients suffering from diabetes require constant monitoring.

he wearable devices can play a major role in connecting patients

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

[

(

ith diabetes to their care teams for effective diabetes manage-

ent [98] . Various technologies have already been developed to

ase diabetes management. For example, Dexcom G6 CGM System

s a smartphone-connected system for constant glucose monitor-

ng (CGM) [99] . Receiving the right dosage of insulin based on the

lucose level is critical. Hence, MiniMed 770G System [100] has

een developed which is an insulin pump delivering appropriate

osage of insulin to the user based on glucose reading. Another

earable used to track the daily insulin intake has been developed

y Companion Medical [101] . The wearable is a smart pen which

onnects to the patient’s phone via Bluetooth. The patient uses the

en to inject insulin. The time of injection and the injected dosage

s sent to the patient’s smartphone by the smart pen. This way the

atient can easily manage the daily intake of insulin. Alfian et al.

102] have used DL to develop a blood glucose smart sensor. To

his end, blood pressure, blood glucose, and heart rate were fed

o a LSTM model for real-time diabetes classification using cloud

ervice.

Hemodynamics is the study of blood flow and researchers have

eveloped a wearable cephalic laser blood flowmeter for inves-

igation of hemodynamics upon changing body posture (e.g. ris-

ng from a sitting posture). The developed device is worn on the

ragus [103] . In another research, site-specific blood flow varia-

ions in people during running were detected using a laser doppler

owmeter which is wearable [104] .

.2.3. Helping with movement disorders

Wearables can also help people with movement disorder. For

xample, Phoenix suit is a wearable exoskeleton which helps in

he movement of knees and hips using small motors. The move-

ent of the suit is controlled by pushing buttons integrated into

he suit. The movement disabilities may have been acquired later

n life. For example, people suffering from stroke may experience

cquired disabilities. Patients with stroke need long-term therapy

o regain their movement abilities. The therapies may be expensive

r even inaccessible due to social and environmental factors. Wear-

ble devices can be used to monitor the patient’s activities and

rovide feedback to the patient and therapist to make home exer-

ise programs possible. An example of such wearable devices was

eveloped by Burridge et al. [105] . Their wearable was equipped

ith embedded inertial and mechanomyographic sensors. The col-

ected data from these sensors were used to classify functional

ovements of the patient to provide useful information. Another

earable device to monitor patient’s exercises was developed by

urns and Adeli [106] . This device can help patients with brain

nd spinal cord injuries to manage their exercise programs to re-

over their movement abilities. The developed wearable records

atient’s physiological data as he/she performs the required exer-

ises. The recorded data are then sent to clinicians from patient’s

ome. The clinicians carry out the necessary supervision based on

he received data from the wearables, remotely.

Even people without disabilities may need help and protection

hen they get old. Falling during walking is common among el-

erly people. While falling is not considered a major risk for young

eople, which may lead to severe injuries for old people. González

t al. [107] used two accelerometers that are worn as bracelets and

mployed genetic algorithm (GA) for fall detection. Pannurat et al.

108] proposed another fall detection attempt using a wireless

earable accelerometer and classification algorithms. Their method

s a combination of a rule-based knowledge representation, a time

ontrol mechanism, and machine-learning-based activity classifica-

ion. The method has been used for fall detection at pre-impacts,

mpacts, and post-impacts, respectively. Another fall detection sys-

em for elderly people based on smartwatch data were proposed

y Mauldin et al. [109] . They used a GRU RNN as the predictive

odel to perform fall detection. The predictive model was de-

loyed on cloud to make real-time decision making.

.2.4. Mental health

Mental health is as important as physical health and re-

earchers have already begun to development of many wearables

or mental condition monitoring. These wearables can usually de-

ermine human physiology status based on collected data such as

eartbeat, blood pressure, body temperature, or ECG. One typical

pplication of wearables is stress assessment. Choi et al. [110] col-

ected heart rate and audio signals from children using wearable

evices. These data coupled with support vector machine (SVM)

ere used to detect the stress patterns in children. Emotion board

s another attempt made for stress detection [111] based on elec-

rodermal activity (EDA). In this project, the collected skin con-

uctance signals were processed using linear discriminant analysis

LDA) and classified using SVM.

Wearable technologies can also be helpful in diagnosing and

onitoring of psychiatric disorders such as depression. Valenza

t al. [112] used PHYCE system to collect data for assessment of

he depressive status in bipolar disorder. PHYCE is a wearable sys-

em prototype that detects the ECG using textile electrodes and ac-

uires the respiration signal using piezoresistive sensors. In other

esearch, Roh et al. [113] developed a system-on-chip (SoC) to

ccelerate filtering and feature extraction of heart-rate variability

HRV) from an ECG. They managed to improve the accuracy of de-

ression recognition.

.2.5. Autism

Children with autism spectrum disorder suffer from emotion

ecognition deficits. Therefore, they need help to improve their

motion recognition abilities. Daniels et al. [114] developed a pro-

otype therapeutic tool using Google Glass for autistic children.

hey reported that autistic children had no problem in wearing

he device. In their work, set of images illustrating different emo-

ional states were shown to autistic children. Showing the correct

motional classification of images to the children via Google Glass

mproved their emotion recognition abilities.

Another application using machine learning and wearable tech-

ology is the detection of stereotypical motor movements (SMMs)

ased on real-time measurements from IMU which are sent for

rocessing to a cloud service [115] . SMMs are associated with

utism spectrum disorders. The proposed approach consisted of

wo phases namely feature extraction and decision making. A CNN

as used for feature extraction and the extracted features were fed

o a LSTM for decision making.

.2.6. Healthcare wearables shortcomings

Although wearable devices have considerable potential applica-

ions in healthcare domain, they have several shortcomings that

ust be addressed. For example, most of wearable devices are still

n the prototype stage and need extensive evaluation before being

ualified as final products. Wearable devices may connect to cloud

ervices to process and store the data collected by them. Therefore,

nforcing privacy of patients’ medical data is crucial. As the num-

er of wearable devices grows, the amount of data generated by

hem grows as well. Thus big data can be considered both a con-

ern and an opportunity for artificial intelligence research commu-

ity.

.3. Industrial and manufacturing

As industrial infrastructures evolve, performing the desired

asks with efficiency, accuracy and speed is highly desirable. With

ufficient research and development, wearable devices can have

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

(

[

(

he potential to revolutionize the modern industry. Whether hu-

ans can be totally replaced by machines in the future, is al-

ays debatable. The machines will take the humans’ place in doing

epetitive and routine tasks. However, the total removal of human

upervision is not likely in cases that human’s experience is re-

uired [116] . Currently, wearables have gained considerable share

f consumer market. However, application of wearables in the in-

ustry is still limited.

.3.1. Examples of real-world industrial wearables

There have been several research attempts to pave the road for

pplication of wearables in the industry. In WearIT@work project,

earable computing group has conducted research in application

f wearables in production domain [116] . The research was carried

ut in Skoda Auto car manufacturing. The objective of the research

as to replace the traditional paper-based car quality assurance

erformed at the end of the assembly line. A wearable device con-

isting of a belt-computer and a head-mounted display was devel-

ped in a way that the workers need not be trained in order to

se it. The main achievement of this work was the identification

f suitable methods for integrating wearable devices in real-life in-

ustrial scenarios. Some of unaddressed challenges of the proposed

earable were battery life for multiple working shifts and lack of

eliable wireless communication at the user level.

Wearable computer systems group at Carnegie Mellon Univer-

ity (CMU) has conducted another research entitled Navigator 2

bout industrial applications of wearables [116] . The aim of this

roject was improvement of inspection routines of mobile workers.

efore using the developed wearable, the workers had to complete

checklist with hundreds of pages. The checklist completion took

our to six hours. Performing the same inspection routine but with

he help of the developed wearable reduced the inspection time up

o 50%. The proposed wearable had support for speech recognition

eading to minimum interference of the wearable with the worker

uring inspection. The tackled challenges during this research were

nterface design, cognitive model, contextual awareness, and adap-

ation to tasks being performed. The factors left for further inves-

igation were weight limit of the wearable and its long-term effect

n the wearer’s body.

Efficient order picking is critical to maximize the gain of any

anufacturing line. Contextual computing group at Georgia Insti-

ute of Technology (Georgia Tech) has conducted a research to in-

estigate the effect of using wearables for order picking [116] . The

esearch was based on Google Glass which is a head-up display

HUD). The experimental results showed that using HUD to aid

he workers with order picking task to reduce the possible hu-

an errors and part picking time. The unaddressed challenges of

his research were wearability of head-mounted displays and opti-

ization of decision making for mobile workers through wearable

omputing.

.3.2. Critical design factors for industrial wearables

An important stepping stone towards making wearables prac-

ical in industrial applications is careful inspection of problems

f using wearables in the industry. To this end, in [116] , 25 en-

erprises were surveyed based on four aspects: 1. industrial sec-

or, 2. application scenarios, 3. current data processing methods,

nd 4. data interaction level. The industrial sectors investigated

y the survey were equipment manufacturing, metallurgy, chem-

cal, warehousing, rail transit, airport, and harbor. The considered

pplication scenarios were manufacturing execution, equipment

anagement, order picking, remote assistance, asset management

nd warehouse operation. Manufacturing execution tracks and doc-

ments the transformation of raw materials into final products

117] . Equipment management revolves around checking status of

arge-scale facilities. Order picking is one of the common opera-

ions in e-commerce warehousing. Remote assistance is related to

he application of augmented reality (AR) glasses. Asset manage-

ent is related to checking inventory for non-production items.

he data processing in the selected 25 enterprises was carried out

sing paper, personal digital assistant (PDA) or personal computer

PC).

Based on the four aspects (industrial sector, application scenar-

os, data processing method, and data interaction level), the sur-

eyed enterprises are partitioned and the results are shown in (a)-

d) of Fig. 7 .

To identify the key challenges and shortcomings of applying

earables in the industry, in the 25 enterprises, multiple wear-

ble devices have been tested by several users. Based on the re-

eived feedback from the users, five critical factors were deter-

ined which must be taken into account for designing practical

earables for industrial applications:

1. Ergonomic product design:

a. The wearables must be lightweight especially if they are

head-mounted.

b. The parts of the wearables that make contact with human

skin must be made of comfortable materials.

c. Wearables must provide hand-free experience for the users

since they need to perform various tasks with their hands.

Occupying the user’s hands or restricting their movements

due to wired connections is not an option.

2. Data interaction on device:

a. The wearables should provide the key information concisely.

Showing too much information on the wearable screen may

disturb the user.

b. Limiting use of touch screen and keyboard for receiving in-

put from users. This factor is due to the fact that wearables

usually do not support complicated data typing.

c. Voice and gesture interaction may be used as supplemen-

tary interaction methods.

3. Operational stability:

a. The industrial wearables should be equipped with batter-

ies that can last for more than eight hours (one work-shift)

without charging.

b. The stability of network connections such as Wifi, Blue-

tooth, etc. is critical in industrial environments. Moreover,

the network connections should be easily deployable and

they should be easy-to-use.

c. Industrial environments are usually harsh with high temper-

ature, humidity, and shock. The industrial wearables must

stay operational under such conditions.

4. External software integration:

a. Considering that industrial wearables cannot work indepen-

dently, they are required to be integrated into the enterprise

systems seamlessly.

b. The industrial wearables must have the ability to process

data in real-time and make-decentralized decisions.

c. The wearables must allow the involvement of human expe-

rience whenever needed.

5. External hardware integration:

a. The wearables must be able to accept/collect data from ma-

chines or robots of the manufacturing site.

b. The wearables must be able to control external equipment.

Such requirement provides the user to control the equip-

ment and intervene with the operation process if needed.

c. The wearables must support human-machine cooperation.

This way the flexibility of the human and the accuracy of

the machines can be combined to improve efficiency. Using

the wearable, the human can instruct the machines with a

series of gestures and signs.

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

[

(

Three wearable projects are reviewed in Section 7.3.1 based on

he five design factors presented above. The evaluation results are

resented in Fig. 8 . Apparently, all three projects have tried to de-

ign their wearable devices in an ergonomic way. Wearable com-

uting group has the best operational stability, whereas contextual

omputing group has the best data interaction on device.

.4. Human–robot interaction

Human-robot interaction (HRI) is about establishing efficient,

afe, and comfortable interactions between humans and robots.

he interaction usually takes place via a wearable medium. That

s where wearable devices come in. One of these wearables is a

rist-worn camera called WristCam [118] which is designed for

and gesture recognition. This wearable relies on speeded up ro-

ust features (SURF) [119] which are matched between successive

rames of video captured by the camera. The user’s hand velocity

s determined using feature matching. After hand gesture extrac-

ion, it is segmented based on a predefined gesture starting signal.

he gesture segments are then classified using the dynamic time

arping (DTW) [120] method. In addition to robot control based

n hand gesture recognition, it is also possible to command a robot

ased on the walking pattern. Cifuentes et al. [121] proposed a hu-

an tracking approach for a service robot using a wearable IMU

nd laser ranger finder mounted on the robot. IMU was used to

apture the walking pattern of the human and laser range finder

as used to detect the human’s legs. Based on the sensed data,

he human tracking system was able to control the robot such that

t followed the human walking pattern. The tracking system was

valuated in an eight-shaped trajectory.

One of the approaches to realizing HRI is skill learning. Fang

t al. [122] proposed a skill learning approach for HRI using a

earable device. Their proposed system consists of two subsys-

ems: 1. Robot teleoperation, and 2. Imitation learning. The teleop-

ration is implemented using the robotic operating system (ROS)

123] and is used to collect training data for imitation learning.

mitation learning relies on dynamic movement primitive (DMP)

124–126] to mimic the trajectory demonstrated by the user dur-

ng the teleoperation phase. To this end, the user’s arm and hand

otion are recorded using a wearable device equipped with mul-

iple inertial measurements and magnetic units (IMMUs).

Physical human-robot interaction (pHRI) can be used for re-

abilitation, assistive devices, etc. Wearable devices that are used

or pHRI must preserve their users’ comfort and safety. Ghonasgi

t al. [127] proposed a modular sensing panel for pHRI which has

he ability to capture the fine nature of force transmission from

ompliant human tissue onto rigid surfaces in the wearable de-

ice. Their sensing panel uses force-sensing resistors (FSRs) and it

s low-cost and can be adapted to a variety of human interfaces.

nother work regarding pHRI was presented by Lenzi et al. [128] .

he authors focus on a distributed approach for monitoring physi-

al interaction between a user and a wearable robot. To this end, a

istributed tactile sensor consisting of a matrix of optoelectronic

ensors is used. The sensors are embedded in a thin and com-

liant silicone bulk onto the user-robot contact surface. While the

actile sensor is capable of measuring the pressure distribution on

he wearable-human interaction area, it preserves the user’s safety

nd comfort and does not put any specific design constraint on the

obot to house it.

. Wearable technologies challenges

Nowadays, wearable devices are often available in the form of

martwatches which can connect with smartphones. In the future,

earables are expected to be seen in various forms designed for

ifferent applications. The world of the future is the world of wear-

ble devices that can help the humankind in doing his duties. They

an market in the fastest possible time by sharing the collected

ata and help to maximize the profit. It can be generalized to other

spects of life like in medical, geographical, or personal fields. The

ollected data by wearables in the form of text, video, audio or

ther specified forms can be shared to help with accurate disease

iagnosis.

The current generation of wearable devices is still far from per-

ect. The developed technologies are impressive albeit not mature

nough. To unleash the full power of wearable devices, multiple

hallenges must be addressed. Some of the challenges faced by the

earable devices are briefly discussed in the sections below.

.1. Data collection

The first challenge is related to data acquisition. The qual-

ty, quantity, resolution, and other parameters of the gathered

ata depend on the wearable device. Spatial resolution, tempo-

al resolution, or data resolution are the factors which may im-

act data quality and quantity [129] . Collecting data from users

n an optimal manner is challenging. The gathered raw data must

e pre-processed before its clinical application. To this end, the

easured quantities from different devices must be unified and

heir errors and statistical outliers must be removed. After being

re-processed, the data are ready to be used by data analytics.

ata processing solutions for wearable data often rely on machine

earning [ 130 , 131 ]. Obtaining high quality labels for the data is

ime consuming and requires expert knowledge or intervention of

he wearable user [ 132 , 133 ]. Wearables that require insertion into

he user’s body like insertable cardiac monitors, continuous glu-

ose monitors, and insertable drug deliverables systems have their

wn challenges:

• Foreign body reaction may impede the functioning of biosen-

sors and their data transmission. • Inserted device might move unexpectedly.

.2. Data transmission

Coming up with an energy-efficient solution to transmit data

collected by wearables) for further processing is crucial. The emer-

ence of faster connection technologies such as 5G and beyond

eads to ever-increasing amount of data generation. Processing and

torage of these data is challenging. Relying solely on centralized

loud computing is not an option due to data processing latency

nd significant load on network performance. Edge computing can

educe latency by moving the necessary computation on the net-

ork’s edge. However, there are still issues with development of

oftware and hardware of edge devices which must be fixed to

eet the cloud computing load [134] .

.3. Security and privacy

Enforcing privacy, security, and trustworthiness while using

earables is still an open challenge [135] . The main feature of the

earables is continuous sensing and data collection. As mentioned

n [136] , most modern wearables can collect data about position,

hysical activity level, and mental health of users who wear them.

rom the user’s point of view, these data might be considered sen-

itive, so enforcing their privacy cannot be overlooked. Currently,

here is no unified solution to address all the potential security

nd privacy threats of wearables so more research and develop-

ent are required to improve the security and privacy aspects of

earable devices.

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

[

.4. Localization quality

In many applications of wearable technologies, precise localiza-

ion of the wearable devices is important. Given that wearables are

sually resource-constrained, achieving localization with accept-

ble precision is challenging. Therefore, improving the localization

uality of wearable devices under limited computational power is

eeded.

.5. User adoption aspects

The success of wearable technologies directly depends on how

uch the target users would accept to use them. User adoption

s specifically challenging in medical and industrial applications. In

ll other applications, user adoption is a matter of personal choice.

owever, in medical and industrial domains, using wearables is

ore of a necessity than a choice. In the medical domain, the pa-

ients may feel discomfort and stress about wearing pervasive de-

ices. This is mainly due to the complexity and excessive ‘‘intru-

iveness’’ of these devices. In the industrial domain, some work-

rs may fail to understand the benefit and purpose of monitoring

earable devices and may resist using them.

.6. Resource constraints

Providing new services and targeting new users require the

evelopment of advanced functionalities for the wearables. How-

ver, adding new functionalities increases the power consumption

f already resource-constrained wearables. Sometimes, the quality

f the final wearable product is not met due to limited resource

equirements. Therefore, managing energy consumption and yet

chieving the expected performance is one of the most important

hallenges of the wearables.

.7. Interoperability

In the internet of wearable things (IoWT), different wearable

evices must be able to communicate with each other regardless

f their technologies. Such device-to-device (D2D) communication

etween wearable devices with different com putational power (e.g.

ow-end and high-end devices) is a stepping-stone toward the real-

zation of various smart functions in a decentralized manner. Recall

hat an individual wearable device does not have much to offer due

o its limited resources. However, with efficient management and

2D communication, the processing power of multiple wearables

an be combined to achieve the complex tasks. Currently, reliable

2D communication is one of the open problems of wearable tech-

ologies. Moreover, to fully benefit from the IoWT, developing end-

o-end solutions to achieve seamless integration of wearable things

nto existing systems is one of the great concerns.

. Discussion

Wearables provide various monitoring and scanning features

uch as biofeedback or other sensory physiological functions like

iometry-related ones [137] . Moreover, wearables are portable and

an be used hands-free. Wearable devices may improve life quality

ignificantly but first, they have to be cost-effective.

According to [138] , about half of people who purchase a wear-

ble stop using it. One-third of them do this before six months.

s reported in [139] , elderly people have shown interest in using

earable devices for physical and mental health purposes. How-

ver, due to the lack of awareness about wearable technologies,

urrently many elderly people do not use wearables. Thus, peo-

le need to be trained about the working mechanism of wearable

evices and their maintenance.

The design and development of wearable devices must take into

ccount user preferences. This is particularly important when the

evices have to be worn for longer durations, for example during

hronic diseases monitoring, or data collection about the user’s ac-

ivity level [140] . Being lightweight is one of the important user

references about wearables which severely limits the battery ca-

acity of wearable devices. Therefore, computational power and

ireless communication of wearables will be limited.

The sensors in wearable devices generate lots of data while

alking or jogging. These data can be used to discover dominant

atterns among the population. Researchers of nursing science

ave already taken interdisciplinary approaches to study the med-

cal problems based on big data collected from a huge population.

o this end, multiple professionals with complementary expertise

ave worked as a team [141] . Moreover, the growth of IoWT in-

reases the complexity and amount of generated data. These data

an be used for implementing IoT sensing-based health monitoring

nd management [142] and developing mobile health applications

143] .

As described in Section 8.7 , interoperability is one of the chal-

enges of wearable technologies which is an important direction

or future works; it is also an important requirement for remote

ealthcare services. The fifth generation of wireless networking

echnology (5G) allows the connection of many hospital devices

o the network and provides remote access from home. An exam-

le of remote healthcare using wearables is the Hospital Without

alls project developed by Australia’s Commonwealth Scientific

nd Industrial Research Organization (CSIRO). This project aims

o monitor patients continuously in certain diagnostic categories

144] . In this project, a miniature, wearable, and low-power radio

s used to transmit vital signs and activity information to a home

omputer. The data are then sent by telephone line or through

he internet to appropriate medical experts. Another important fu-

ure trend is the emergence of new wearable devices. For example,

n the medical domain, it is expected that drug delivery systems

ill emerge in the form of wearable devices (e.g. MiniMed 770G

n Section 7.2.2 .). Disease intervention by wearables is also ex-

ected via integration with actuators planted insides/on the body.

I methods have already attracted the attention of researchers of

earable devices. Authors in [145] presented a proof-of-concept

or a seizure prediction system. This system utilized a deep learn-

ng classifier to distinguish between preictal and interictal EEG sig-

als. The deep learning model in combination with neuromorphic

ardware formed a wearable seizure warning system which is suit-

ble for patient-specific settings.

0. Conclusion

Wearable technology is an essential building block in future in-

ormation and communication technology (ICT) systems. However,

earable technology has not reached an acceptable level of matu-

ity yet. Multiple challenges are still unaddressed with regard to

ata collection, data processing, communications, security, etc. The

im of this review was to give readers a broad overview of applica-

ions of wearable devices in sport, medical, and industrial domains.

n the future, it will be useful to also investigate the application of

earables in other domains. In this review, the role of AI meth-

ds in the development of wearable devices has been investigated

s well. As future research, further applications of AI techniques

n wearable devices to improve the quality of life by monitoring

hysiological parameters or early automated detection of diseases

an be investigated.

unding

There is not funding source for this paper.

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

eclaration of Competing Interest

The authors have no conflicting interests to declare.

eferences

[1] A. Ometov, et al., A survey on wearable technology: history, state-of-the- art and current challenges, Comput. Netw. 193 (2021) 108074, doi: 10.1016/

j.comnet.2021.108074 . [2] N. Ayoobi , et al. , Time series forecasting of new cases and new deaths rate for

COVID-19 using deep learning methods, Results in Physics 27 (2021) 1–15 . [3] H. Asgharnezhad et al., “Objective evaluation of deep uncertainty predictions

for covid-19 detection,” arXiv preprint arXiv: 2012.11840 , (2020) 1–7.

[4] R. Alizadehsani , et al. , A data mining approach for diagnosis of coronary artery disease, Comput. Methods Programs Biomed. 111 (1) (2013) 52–61 .

[5] B. Handwerk , With wearable devices that monitor air quality, scientists can crowdsource pollution maps, Smithsonian Mag. (2015) .

[6] Y. Qiu , Security analysis for the information of wearable devices, Netinfo Se- cur. 9 (2016) 79–83 .

[7] J. Twentyman , Wearable devices aim to reduce workplace accidents, Financ.

Times (June 2016) . [8] S. Seneviratne , et al. , A survey of wearable devices and challenges, IEEE Com-

mun. Surv. Tutor. 19 (4) (2017) 2573–2620 . [9] A. Banerjee , C. Chakraborty , A. Kumar , D. Biswas , Chapter 5 - Emerging trends

in IoT and big data analytics for biomedical and health care technologies, in: V.E. Balas, V.K. Solanki, R. Kumar, M. Khari (Eds.), Handbook of Data Science

Approaches for Biomedical Engineering, Academic Press, 2020, pp. 121–152 .

[10] A.K. Yetisen, J.L. Martinez-Hurtado, B. Ünal, A. Khademhosseini, H. Butt, Wear- ables in medicine, Adv. Mater. 30 (33) (2018) 1706910, doi: 10.1002/adma.

201706910 . [11] W. Barfield , Fundamentals of Wearable Computers and Augmented Reality,

CRC press, 2015 . [12] G. Gaikwad , A. Jain , Feelbot: reducing use of bad words in children through

wearable using artificial intelligence and gamification, in: Proceedings of the 2017 Conference on Interaction Design and Children, 2017, pp. 777–781 .

[13] K. Shameer, M.A. Badgeley, R. Miotto, B.S. Glicksberg, J.W. Morgan, J.T. Dudley,

Translational bioinformatics in the era of real-time biomedical, health care and wellness data streams, Brief. Bioinform. 18 (1) (2017) 105–124, doi: 10.

1093/bib/bbv118 . [14] S. Patel , H. Park , P. Bonato , L. Chan , M. Rodgers , A review of wearable sen-

sors and systems with application in rehabilitation, J. Neuroeng. Rehabil. 9 (1) (2012) 1–17 .

[15] J.A. Levine , A five-step strategy to combine data sources from multiple wear-

able sensors, Technol. Investm. 8 (1) (2017) 33–43 . [16] J. Wei, How wearables intersect with the cloud and the internet of things:

considerations for the developers of wearables, IEEE Consum. Electron. Mag. 3 (3) (2014) 53–56, doi: 10.1109/MCE.2014.2317895 .

[17] Y. Athavale, S. Krishnan, Biosignal monitoring using wearables: observations and opportunities, Biomed. Signal. Process Control 38 (2017) 22–33, doi: 10.

1016/j.bspc.2017.03.011 .

[18] M.E. Chowdhury , A. Khandakar , Y. Qiblawey , M.B.I. Reaz , M.T. Islam , F. Touati , Machine learning in wearable biomedical systems, Sports Sci. Hum. Health-D-

iff. Approach. : IntechOpen (2020) . [19] U.R. Acharya, et al., A deep convolutional neural network model to classify

heartbeats, Comput. Biol. Med. 89 (2017) 389–396, doi: 10.1016/j.compbiomed. 2017.08.022 .

[20] B.N. Singh, A.K. Tiwari, Optimal selection of wavelet basis function applied to

ECG signal denoising, Digit Signal Process 16 (3) (2006) 275–287, doi: 10.1016/ j.dsp.20 05.12.0 03 .

[21] Y. Lin, et al., Artificial intelligence of things wearable system for cardiac dis- ease detection, in: 2019 IEEE International Conference on Artificial Intelli-

gence Circuits and Systems (AICAS), 18-20, March 2019, pp. 67–70, doi: 10. 1109/AICAS.2019.8771630 . 2019 .

[22] J. Chan, A.C. Miller, and E.B. Fox, “Representing and denoising wearable ECG

recordings,” arXiv preprint arXiv: 2012.00110 , 2020. [23] G. Lee, B. Choi, H. Jebelli, C.R. Ahn, S. Lee, Noise reference signal–based de-

noising method for EDA collected by multimodal biosensor wearable in the field, J. Comput. Civil Eng. 34 (6) (2020) 04020044, doi: 10.1061/(ASCE)CP.

1943-5487.0 0 0 0927 . [24] M.B. Hossain, J. Lázaro, Y. Noh, K.H. Chon, Denoising wearable armband

ECG data using the variable frequency complex demodulation technique,

in: 2020 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), 20-24, July 2020, pp. 592–595, doi: 10.

1109/EMBC44109.2020.9175665 . 2020 . [25] A. Shoeibi et al., "Automated detection and forecasting of covid-19 using deep

learning techniques: a review," arXiv preprint arXiv: 2007.10785 , 2020. [26] H.T. Tor , et al. , Automated detection of conduct disorder and attention deficit

hyperactivity disorder using decomposition and nonlinear techniques with EEG signals, Comput. Methods Programs Biomed. 200 (2021) 105941 .

[27] A. Shoeibi, et al., Epileptic seizures detection using deep learning techniques:

a review, Int. J. Environ. Res. Public Health 18 (11) (2021) 5780 [Online]. Avail- able https://www.mdpi.com/1660-4601/18/11/5780 .

[28] G.H. Dunteman , Principal Components Analysis, 69, Sage, 1989 . [29] S. Balakrishnama , A. Ganapathiraju , Linear discriminant analysis-a brief tuto-

rial, Inst. Signal Inf. Process. 18 (1998) 1–8 1998 .

[30] P. Comon, Independent component analysis, A new concept? Signal Process. 36 (3) (1994) 287–314, doi: 10.1016/0165- 1684(94)90029- 9 .

[31] S.T. Roweis, L.K. Saul, Nonlinear dimensionality reduction by locally linear embedding, Science 290 (5500) (2000) 2323, doi: 10.1126/science.290.5500.

2323 . [32] F. Khozeimeh , et al. , Combining a convolutional neural network with autoen-

coders to predict the survival chance of COVID-19 patients, Sci. Rep. 11 (1) (2021) 1–18 .

[33] A. Ng, “Machine learning and AI via brain simulations,” Accessed: May, vol. 3,

p. 2018, 2013. [34] R. Alizadehsani, et al., Handling of uncertainty in medical data using machine

learning and probability theory techniques: a review of 30 years (1991–2020), Ann. Oper. Res. 2 (2021) 1–42, doi: 10.1007/s10479- 021- 04006- 2 .

[35] R. Caruana , A. Niculescu-Mizil , An empirical comparison of supervised learn- ing algorithms, in: Proceedings of the 23rd International Conference on Ma-

chine Learning, 2006, pp. 161–168 .

[36] R. Alizadehsani, et al., Model uncertainty quantification for diagnosis of each main coronary artery stenosis, Soft Comput. 24 (13) (2020) 10149–10160,

doi: 10.10 07/s0 050 0- 019- 04531- 0 . [37] M. Khodatars , et al. , Deep learning for neuroimaging-based diagnosis and re-

habilitation of autism spectrum disorder: a review, Comput. Biol. Med. 139 (2021) 104 94 9 .

[38] A. Shoeibi et al., “Applications of deep learning techniques for automated

multiple sclerosis detection using magnetic resonance imaging: a review,” arXiv preprint arXiv: 2105.04881 , 2021.

[39] Z. Arabasadi , R. Alizadehsani , M. Roshanzamir , H. Moosaei , A .A . Yarifard , Com- puter aided decision making for heart disease detection using hybrid neural

network-Genetic algorithm, Comput. Methods Programs Biomed. 141 (2017) 19–26 .

[40] E. Nasarian, et al., Association between work-related features and coronary

artery disease: a heterogeneous hybrid feature selection integrated with bal- ancing approach, Pattern Recognit. Lett. 133 (2020) 33–40, doi: 10.1016/j.

patrec.2020.02.010 . [41] D. Sharifrazi , et al. , Fusion of convolution neural network, support vec-

tor machine and Sobel filter for accurate detection of COVID-19 pa- tients using X-ray images, Biomed. Signal. Process Control 68 (2021)

102622 .

[42] S.L. Oh , et al. , Classification of heart sound signals using a novel deep WaveNet model, Comput. Methods Programs Biomed. 196 (2020) 105604 .

[43] R. Alizadehsani et al., "Uncertainty-aware semi-supervised method using large unlabelled and limited labeled COVID-19 data," arXiv preprint arXiv: 2102.

06388 , 2021. [44] M. Khodatars et al., "Deep learning for neuroimaging-based diagnosis

and rehabilitation of autism spectrum disorder: a review," arXiv preprint

arXiv: 2007.01285 , 2020. [45] R. Alizadehsani , et al. , Coronary artery disease detection using artificial intelli-

gence techniques: a survey of trends, geographical differences and diagnostic features 1991–2020, Comput. Biol. Med. 128 (2021) 104095 .

[46] M.A. Fahami, M. Roshanzamir, N.H. Izadi, V. Keyvani, R. Alizadehsani, Detec- tion of effective genes in colon cancer: a machine learning approach, Inform.

Med. Unlocked 24 (2021) 100605, doi: 10.1016/j.imu.2021.100605 . [47] X. Zhu , A.B. Goldberg , Introduction to semi-supervised learning, Synth. Lect.

Artif. Intell. Machine Learn. 3 (1) (2009) 1–130 .

[48] By A.G. Barto , R.S. Sutton , Reinforcement learning: an introduction, SIAM Rev. 63 (2) (June 2021) 423 .

[49] S. Saadatnejad, M. Oveisi, M. Hashemi, LSTM-based ECG classification for con- tinuous monitoring on personal wearable devices, IEEE J. Biomed. Health In-

form. 24 (2) (2020) 515–523, doi: 10.1109/JBHI.2019.2911367 . [50] A. Amirshahi, M. Hashemi, ECG classification algorithm based on STDP and R-

STDP neural networks for real-time monitoring on ultra low-power personal

wearable devices, IEEE Trans. Biomed. Circuits Syst. 13 (6) (2019) 1483–1493, doi: 10.1109/TBCAS.2019.2948920 .

[51] J. Acharya, A. Basu, Deep neural network for respiratory sound classifica- tion in wearable devices enabled by patient specific model tuning, IEEE

Trans. Biomed. Circuits Syst. 14 (3) (2020) 535–544, doi: 10.1109/TBCAS.2020. 2981172 .

[52] M.D. Hssayeni, J. Jimenez-Shahed, M.A. Burack, B. Ghoraani, Wearable sen-

sors for estimation of Parkinsonian tremor severity during free body move- ments, Sensors 19 (19) (2019) 4215 [Online] Available https://www.mdpi.com/

1424-8220/19/19/4215 . [53] C. Ahlrichs, et al., Detecting freezing of gait with a tri-axial accelerometer in

Parkinson’s disease patients, Med. Biol. Eng. Comput. 54 (1) (2016) 223–233, doi: 10.1007/s11517-015-1395-3 .

[54] R. Varatharajan, G. Manogaran, M.K. Priyan, R. Sundarasekar, Wearable sen-

sor devices for early detection of Alzheimer disease using dynamic time warping algorithm, Cluster Comput. 21 (1) (2018) 681–690, doi: 10.1007/

s10586- 017- 0977- 2 . [55] A. Das, et al., Unsupervised heart-rate estimation in wearables with Liquid

states and a probabilistic readout, Neural Netw. 99 (2018) 134–147, doi: 10. 1016/j.neunet.2017.12.015 .

[56] A. Krause , D.P. Siewiorek , A. Smailagic , J. Farringdon , Unsupervised, Dynamic

Identification of Physiological and Activity Context in Wearable Computing, in: Unsupervised, Dynamic Identification of Physiological and Activity Context

in Wearable Computing, 3, ISWC, 2003, p. 88 . [57] R. Janarthanan, S. Doss, S. Baskar, Optimized unsupervised deep learning as-

sisted reconstructed coder in the on-nodule wearable sensor for human activ-

https://doi.org/10.1016/j.comnet.2021.108074

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0002

http://arxiv.org/abs/arXiv:2012.11840

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0004

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0005

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0006

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0007

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0008

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0009

https://doi.org/10.1002/adma.201706910

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0011

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0012

https://doi.org/10.1093/bib/bbv118

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0014

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0015

https://doi.org/10.1109/MCE.2014.2317895

https://doi.org/10.1016/j.bspc.2017.03.011

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0018

https://doi.org/10.1016/j.compbiomed.2017.08.022

https://doi.org/10.1016/j.dsp.2005.12.003

https://doi.org/10.1109/AICAS.2019.8771630

http://arxiv.org/abs/arXiv:2012.00110

https://doi.org/10.1061/(ASCE)CP.1943-5487.0000927

https://doi.org/10.1109/EMBC44109.2020.9175665

http://arxiv.org/abs/arXiv:2007.10785

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0026

https://www.mdpi.com/1660-4601/18/11/5780

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0028

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0029

https://doi.org/10.1016/0165-1684(94)90029-9

https://doi.org/10.1126/science.290.5500.2323

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0032

https://doi.org/10.1007/s10479-021-04006-2

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0035

https://doi.org/10.1007/s00500-019-04531-0

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0037

http://arxiv.org/abs/arXiv:2105.04881

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0039

https://doi.org/10.1016/j.patrec.2020.02.010

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0041

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0042

http://arxiv.org/abs/arXiv:2102.06388

http://arxiv.org/abs/arXiv:2007.01285

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0045

https://doi.org/10.1016/j.imu.2021.100605

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0047

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0048

https://doi.org/10.1109/JBHI.2019.2911367

https://doi.org/10.1109/TBCAS.2019.2948920

https://doi.org/10.1109/TBCAS.2020.2981172

https://www.mdpi.com/1424-8220/19/19/4215

https://doi.org/10.1007/s11517-015-1395-3

https://doi.org/10.1007/s10586-017-0977-2

https://doi.org/10.1016/j.neunet.2017.12.015

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0056

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

[

ity recognition, Measurement 164 (2020) 108050, doi: 10.1016/j.measurement. 2020.108050 .

[58] B. Ballinger , et al. , DeepHeart: semi-supervised sequence learning for cardio- vascular risk prediction, in: Proceedings of the AAAI Conference on Artificial

Intelligence, 32, 2018 . [59] K. Yang, C.R. Ahn, M.C. Vuran, S.S. Aria, Semi-supervised near-miss fall de-

tection for ironworkers with a wearable inertial measurement unit, Autom. Constr. 68 (2016) 194–202, doi: 10.1016/j.autcon.2016.04.007 .

[60] M. Stikic, D. Larlus, B. Schiele, Multi-graph based semi-supervised learning for

activity recognition, in: 2009 International Symposium on Wearable Comput- ers, 4-7, Sept. 2009, pp. 85–92, doi: 10.1109/ISWC.2009.24 . 2009 .

[61] M. Stikic , K. Van Laerhoven , B. Schiele , Exploring semi-supervised and active learning for activity recognition, in: 2008 12th IEEE International Symposium

on Wearable Computers, IEEE, 2008, pp. 81–88 . [62] Y. Ma , H. Ghasemzadeh , Labelforest: non-parametric semi-supervised learning

for activity recognition, in: Proceedings of the AAAI Conference on Artificial

Intelligence, 33, 2019, pp. 4520–4527 . [63] G. Wiechert , et al. , Evolutionary semi-supervised rough categorization of

brain signals from a wearable headband, in: 2016 IEEE Congress on Evolu- tionary Computation (CEC), IEEE, 2016, pp. 3131–3138 .

[64] A. Ahadi-Sarkani , S. Elmalaki , ADAS-RL: adaptive vector scaling reinforcement learning for human-in-the-loop lane departure warning, in: Proceedings of

the First International Workshop on Cyber-Physical-Human System Design

and Implementation, 2021, pp. 13–18 . [65] S. Elmalaki, FaiR-IoT: fairness-aware human-in-the-loop reinforcement learn-

ing for harnessing human variability in personalized IoT, in: presented at the Proceedings of the International Conference on Internet-of-Things De-

sign and Implementation, Charlottesvle, VA , USA , 2021 [Online] Available, doi: 10.1145/3450268.3453525 .

[66] M. Baucum, A. Khojandi, R. Vasudevan, and R. Ramdhani, "Optimizing

patient-specific medication regimen policies using wearable sensors in Parkinson’s disease." (2021) 1-39, doi: 10.13140/RG.2.2.16922.77769 .

[67] R.J. Williams , Simple statistical gradient-following algorithms for connection- ist reinforcement learning, Mach. Learn. 8 (3–4) (1992) 229–256 .

[68] H. Jebelli , M.M. Khalili , S. Hwang , S. Lee , A supervised learning-based con- struction workers’ stress recognition using a wearable electroencephalogra-

phy (EEG) device, Constr. Res. Congress 2018 (2018) 43–53 .

[69] A. Das , F. Catthoor , S. Schaafsma , Heartbeat classification in wearables us- ing multi-layer perceptron and time-frequency joint distribution of ECG, in:

Proceedings of the 2018 IEEE/ACM International Conference on Connected Health: Applications, Systems and Engineering Technologies, 2018, pp. 69–

74 . [70] A . Gupta , A . Al-Anbuky , P. McNair , Activity classification feasibility using

wearables: considerations for hip fracture, J. Sens. Actuator Netw. 7 (4) (2018)

54 . [71] V. Nguyen , H. Kunz , P. Taylor , D. Acosta , Insights into pharmacotherapy man-

agement for Parkinson’s disease patients using wearables activity data, MIE (2018) 156–160 .

[72] Y. Guan , T. Plötz , Ensembles of deep lstm learners for activity recognition using wearables, Proc. ACM Interact., Mobile, Wear. Ubiquit. Technol. 1 (2)

(2017) 1–28 . [73] F. Nasirzadeh , et al. , Physical fatigue detection using entropy analysis of heart

rate signals, Sustainability 12 (7) (2020) 2714 .

[74] D. Borthakur , H. Dubey , N. Constant , L. Mahler , K. Mankodiya , Smart fog: fog computing framework for unsupervised clustering analytics in wearable in-

ternet of things, in: 2017 IEEE Global Conference on Signal and Information Processing (GlobalSIP), IEEE, 2017, pp. 472–476 .

[75] N. Weber , A. Härmä, E.P.D.T. Heskes , Unsupervised Learning in Human Activ- ity Recognition: a First Foray into Clustering Data Gathered from Wearable

Sensors Ph. D. Thesis, Radboud University, Nijmegen, The Netherlands, 2017 .

[76] L.F. Mejia-Ricart , P. Helling , A. Olmsted , Evaluate action primitives for human activity recognition using unsupervised learning approach, in: 2017 12th In-

ternational Conference for Internet Technology and Secured Transactions (IC- ITST), IEEE, 2017, pp. 186–188 .

[77] L. Bai , C. Yeung , C. Efstratiou , M. Chikomo , Motion2Vector: unsupervised learning in human activity recognition using wrist-sensing data, in: Adjunct

Proceedings of the 2019 ACM International Joint Conference on Pervasive and

Ubiquitous Computing and Proceedings of the 2019 ACM International Sym- posium on Wearable Computers, 2019, pp. 537–542 .

[78] D. Trabelsi , S. Mohammed , F. Chamroukhi , L. Oukhellou , Y. Amirat , An unsu- pervised approach for automatic activity recognition based on hidden Markov

model regression, IEEE Trans. Autom. Sci. Eng. 10 (3) (2013) 829–835 . [79] R. Gupta , A. Gupta , R. Aswal , Detection of poor posture using wearable sen-

sors and unsupervised learning, in: 2021 7th International Conference on

Advanced Computing and Communication Systems (ICACCS), 1, IEEE, 2021, pp. 527–531 .

[80] V. Dovgalecs , R. Megret , H. Wannous , Y. Berthoumieu , Semi-supervised learn- ing for location recognition from wearable video, in: 2010 International

Workshop on Content Based Multimedia Indexing (CBMI), IEEE, 2010, pp. 1–6 . [81] M. Lv , L. Chen , T. Chen , G. Chen , Bi-view semi-supervised learning based se-

mantic human activity recognition using accelerometers, IEEE Trans. Mobile

Comput. 17 (9) (2018) 1991–2001 . [82] M. Zeng , T. Yu , X. Wang , L.T. Nguyen , O.J. Mengshoel , I. Lane , Semi-super-

vised convolutional neural networks for human activity recognition, in: 2017 IEEE International Conference on Big Data (Big Data, IEEE, 2017, pp. 522–

529 .

[83] J. Zheng , H. Cao , D. Chen , R. Ansari , K.-.C. Chu , M.-.C. Huang , Designing deep reinforcement learning systems for musculoskeletal modeling and locomo-

tion analysis using wearable sensor feedback, IEEE Sens. J. 20 (16) (2020) 9274–9282 .

[84] T. Yamagata, R. Santos-Rodríguez, R. McConville, and A. Elsts, "Online feature selection for activity recognition using reinforcement learning with multiple

feedback," arXiv preprint arXiv: 1908.06134 , 2019. [85] N. Rhinehart , K.M. Kitani , First-person activity forecasting with online inverse

reinforcement learning, in: Proceedings of the IEEE International Conference

on Computer Vision, 2017, pp. 3696–3705 . [86] J. Watts , A. Khojandi , R. Vasudevan , R. Ramdhani , Optimizing individual-

ized treatment planning for parkinson’s disease using deep reinforcement learning, in: 2020 42nd Annual International Conference of the IEEE En-

gineering in Medicine & Biology Society (EMBC), IEEE, 2020, pp. 5406– 5409 .

[87] G. Bhat , R. Deb , V.V. Chaurasia , H. Shill , U.Y. Ogras , Online human activity

recognition using low-power wearable devices, in: 2018 IEEE/ACM Interna- tional Conference on Computer-Aided Design (ICCAD), IEEE, 2018, pp. 1–8 .

[88] N. Rhinehart , K.M. Kitani , First-person activity forecasting from video with online inverse reinforcement learning, IEEE Trans. Pattern Anal. Mach. Intell.

42 (2) (2018) 304–317 . [89] S.-.T. Chen, S.-.S. Lin, C.-.W. Lan, H.-.Y. Hsu, Design and development of a

wearable device for heat stroke detection, Sensors 18 (1) (2018) 17 [Online].

Available https://www.mdpi.com/1424-8220/18/1/17 . [90] C. Skazalski , R. Whiteley , C. Hansen , R. Bahr , A valid and reliable method to

measure jump-specific training and competition load in elite volleyball play- ers, Scand. J. Med. Sci. Sports 28 (5) (2018) 1578–1585 .

[91] R. Maddison , et al. , Usefulness of wearable cameras as a tool to enhance chronic disease self-management: scoping review, JMIR Mhealth Uhealth 7

(1) (2019) e10371 .

[92] E.E. Dooley , N.M. Golaszewski , J.B. Bartholomew , Estimating accuracy at exer- cise intensities: a comparative study of self-monitoring heart rate and physi-

cal activity wearable devices, JMIR Mhealth Uhealth 5 (3) (2017) e7043 . [93] R.T. Welbi, "Health benefits of wearable tech," [Online] https://www.welbi.co/

single- post/5- health- benefits- of- wearable- tech (Accessed July 3, 2021). [94] E.S. Winokur, M.K. Delano, C.G. Sodini, A wearable cardiac monitor for long-

term data acquisition and analysis, IEEE Trans. Biomed. Eng. 60 (1) (2013)

189–192, doi: 10.1109/TBME.2012.2217958 . [95] Y. Heui-Kyung, et al., Application for the wearable heart activity monitoring

system: analysis of the autonomic function of HRV, in: 2008 30th Annual In- ternational Conference of the IEEE Engineering in Medicine and Biology Soci-

ety, 20-25 Aug., 2008, pp. 1258–1261, doi: 10.1109/IEMBS.2008.4649392 . 2008 . [96] Z. Lin, H. Dai, Y. Xiong, X. Xia, S. Horng, Quantification assessment of bradyki-

nesia in Parkinson’s disease based on a wearable device, in: 2017 39th Annual

International Conference of the IEEE Engineering in Medicine and Biology So- ciety (EMBC), 11-15, July 2017, pp. 803–806, doi: 10.1109/EMBC.2017.8036946 .

2017 . [97] M. Delrobaei, N. Baktash, G. Gilmore, K. McIsaac, M. Jog, Using wearable tech-

nology to generate objective Parkinson’s disease dyskinesia severity score: possibilities for home monitoring, IEEE Trans. Neural Syst. Rehabilit. Eng. 25

(10) (2017) 1853–1863, doi: 10.1109/TNSRE.2017.2690578 . [98] N.D. Heintzman , A digital ecosystem of diabetes data and technology: ser-

vices, systems, and tools enabled by wearables, sensors, and apps, J. Diabetes

Sci. Technol. 10 (1) (2016) 35–41 . [99] " https://www.dexcom.com/g6- cgm- system ."

100] " https://www.medtronicdiabetes.com/home ." [101] " https://www.companionmedical.com/ ."

102] G. Alfian , M. Syafrudin , M.F. Ijaz , M.A. Syaekhoni , N.L. Fitriyani , J. Rhee , A personalized healthcare monitoring system for diabetic patients by utilizing

BLE-based sensors and real-time data processing, Sensors 18 (7) (2018) 2183 .

103] T. Fujikawa , O. Tochikubo , N. Kura , T. Kiyokura , J. Shimada , S. Umemura , Mea- surement of hemodynamics during postural changes using a new wearable

cephalic laser blood flowmeter, Circul. J. 73 (10) (2009) 1950–1955 . 104] W. Iwasaki, H. Nogami, S. Takeuchi, M. Furue, E. Higurashi, R. Sawada, De-

tection of site-specific blood flow variation in humans during running by a wearable laser doppler flowmeter, Sensors 15 (10) (2015) 25507–25519 [On-

line]. Available https://www.mdpi.com/1424-8220/15/10/25507 .

105] J.H. Burridge, et al., Telehealth, wearable sensors, and the internet: will they improve stroke outcomes through increased intensity of therapy, motiva-

tion, and adherence to rehabilitation programs? J. Neurol. Phys. Ther. 41 (2017) [Online]. Available https://journals.lww.com/jnpt/Fulltext/2017/07001/

Telehealth, _ Wearable _ Sensors, _ and _ the _ Internet _ .6.aspx . 106] A. Burns , H. Adeli , Wearable technology for patients with brain and spinal

cord injuries, Rev. Neurosci. 28 (8) (2017) 913–920 .

107] S. González, J. Sedano, J.R. Villar, E. Corchado, Á. Herrero, B. Baruque, Features and models for human activity recognition, Neurocomputing 167 (2015) 52–

60, doi: 10.1016/j.neucom.2015.01.082 . 108] N. Pannurat, S. Thiemjarus, E. Nantajeewarawat, A hybrid temporal reasoning

framework for fall monitoring, IEEE Sens. J. 17 (6) (2017) 1749–1759, doi: 10. 1109/JSEN.2017.2649542 .

109] T.R. Mauldin , M.E. Canby , V. Metsis , A.H. Ngu , C.C. Rivera , SmartFall: a smart-

watch-based fall detection system using deep learning, Sensors 18 (10) (2018) 3363 .

[110] Y. Choi , Y.-.M. Jeon , L. Wang , K. Kim , A biological signal-based stress moni- toring framework for children using wearable devices, Sensors 17 (9) (2017)

1936 .

https://doi.org/10.1016/j.measurement.2020.108050

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0058

https://doi.org/10.1016/j.autcon.2016.04.007

https://doi.org/10.1109/ISWC.2009.24

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0061

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0062

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0063

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0064

https://doi.org/10.1145/3450268.3453525

https://www.doi.org/10.13140/RG.2.2.16922.77769

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0067

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0068

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0069

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0070

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0071

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0072

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0073

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0074

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0075

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0076

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0077

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0078

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0079

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0080

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0081

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0082

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0083

http://arxiv.org/abs/arXiv:1908.06134

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0085

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0086

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0087

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0088

https://www.mdpi.com/1424-8220/18/1/17

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0090

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0091

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0092

https://www.welbi.co/single-post/5-health-benefits-of-wearable-tech

https://doi.org/10.1109/TBME.2012.2217958

https://doi.org/10.1109/IEMBS.2008.4649392

https://doi.org/10.1109/EMBC.2017.8036946

https://doi.org/10.1109/TNSRE.2017.2690578

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0098

https://www.dexcom.com/g6-cgm-system

https://www.medtronicdiabetes.com/home

https://www.companionmedical.com/

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0102

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0103

https://www.mdpi.com/1424-8220/15/10/25507

https://journals.lww.com/jnpt/Fulltext/2017/07001/Telehealth,_Wearable_Sensors,_and_the_Internet_.6.aspx

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0106

https://doi.org/10.1016/j.neucom.2015.01.082

https://doi.org/10.1109/JSEN.2017.2649542

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0109

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0110

D. Nahavandi, R. Alizadehsani, A. Khosravi et al. Computer Methods and Programs in Biomedicine 213 (2022) 106541

[

[ [

[

[111] C. Setz, B. Arnrich, J. Schumm, R.L. Marca, G. Tröster, U. Ehlert, Discriminat- ing stress from cognitive load using a wearable EDA device, IEEE Trans. Inf.

Technol. Biomed. 14 (2) (2010) 410–417, doi: 10.1109/TITB.2009.2036164 . [112] G. Valenza, L. Citi, C. Gentili, A. Lanatá, E.P. Scilingo, R. Barbieri, Characteriza-

tion of depressive states in bipolar patients using wearable textile technology and instantaneous heart rate variability assessment, IEEE J. Biomed. Health

Inform. 19 (1) (2015) 263–274, doi: 10.1109/JBHI.2014.2307584 . [113] T. Roh , S. Hong , H.-.J. Yoo , Wearable depression monitoring system with

heart-rate variability, in: 2014 36th Annual International Conference of the

IEEE Engineering in Medicine and Biology Society, IEEE, 2014, pp. 562–565 . [114] J. Daniels , et al. , Feasibility testing of a wearable behavioral aid for social

learning in children with autism, Appl. Clin. Inform. 9 (01) (2018) 129–140 . [115] N.M. Rad, et al., Deep learning for automatic stereotypical motor movement

detection using wearable sensors in autism spectrum disorders, Signal Pro- cess. 144 (2018) 180–191, doi: 10.1016/j.sigpro.2017.10.011 .

[116] X.T.R. Kong, H. Luo, G.Q. Huang, X. Yang, Industrial wearable system: the

human-centric empowering technology in Industry 4.0, J. Intell. Manuf. 30 (8) (2019) 2853–2869, doi: 10.1007/s10845- 018- 1416- 9 .

[117] M. McClellan , Applying Manufacturing Execution Systems, CRC Press, 1997 . [118] F. Chen , et al. , WristCam: a wearable sensor for hand trajectory gesture recog-

nition and intelligent human–robot interaction, IEEE Sens. J. 19 (19) (2018) 8441–8451 .

[119] H. Bay , A. Ess , T. Tuytelaars , L. Van Gool , Speeded-up robust features (SURF),

Comput. Vis. Image Underst. 110 (3) (2008) 346–359 . 120] M. Müller, Dynamic time warping, Information Retrieval for Music and Mo-

tion (Springer journal) 3 (2007) 69–84, doi: 10.1007/978- 3- 540- 74048- 3 _ 4 . [121] C.A . Cifuentes , A . Frizera , R. Carelli , T. Bastos , Human–robot interaction based

on wearable IMU sensor and laser range finder, Rob. Auton. Syst. 62 (10) (2014) 1425–1439 .

122] B. Fang , X. Wei , F. Sun , H. Huang , Y. Yu , H. Liu , Skill learning for hu-

man-robot interaction using wearable device, Tsinghua Sci. Technol. 24 (6) (2019) 654–662 .

123] " https://www.ros.org/ ," 2021. 124] S. Schaal , Dynamic movement primitives-a framework for motor control in

humans and humanoid robotics, in: Adaptive Motion of Animals and Ma- chines, Springer, 2006, pp. 261–280 .

125] A.J. Ijspeert , J. Nakanishi , H. Hoffmann , P. Pastor , S. Schaal , Dynamical move-

ment primitives: learning attractor models for motor behaviors, Neural Com- put. 25 (2) (2013) 328–373 .

126] A.J. Ijspeert , J. Nakanishi , S. Schaal , Movement imitation with nonlinear dy- namical systems in humanoid robots, in: Proceedings 2002 IEEE International

Conference on Robotics and Automation (Cat. No. 02CH37292), 2, IEEE, 2002, pp. 1398–1403 .

127] K. Ghonasgi , S.N. Yousaf , P. Esmatloo , A.D. Deshpande , A modular design for

distributed measurement of human–robot interaction forces in wearable de- vices, Sensors 21 (4) (2021) 1445 .

128] T. Lenzi , et al. , Measuring human–robot interaction on wearable robots: a dis- tributed approach, Mechatronics 21 (6) (2011) 1123–1131 .

129] D. Pal, A. Tassanaviboon, C. Arpnikanondt, B. Papasratorn, Quality of experi- ence of smart-wearables: from fitness-bands to smartwatches, IEEE Consum.

Electron. Mag. 9 (1) (2020) 49–53, doi: 10.1109/MCE.2019.2941462 . 130] D. Ravì, C. Wong, B. Lo, G. Yang, A deep learning approach to on-node sensor

data analytics for mobile or wearable devices, IEEE J. Biomed. Health Inform. 21 (1) (2017) 56–64, doi: 10.1109/JBHI.2016.2633287 .

[131] A. Harris, H. True, Z. Hu, J. Cho, N. Fell, M. Sartipi, Fall recognition using wearable technologies and machine learning algorithms, in: 2016 IEEE In-

ternational Conference on Big Data (Big Data) , 5-8 Dec, 2016, pp. 3974–3976,

doi: 10.1109/BigData.2016.7841080 . 2016 . 132] J. Ker, L. Wang, J. Rao, T. Lim, Deep learning applications in medical image

analysis, IEEE Access 6 (2018) 9375–9389, doi: 10.1109/ACCESS.2017.2788044 . 133] O.D. Lara, M.A. Labrador, A survey on human activity recognition using wear-

able sensors, IEEE Commun. Surv. Tutor. 15 (3) (2013) 1192–1209, doi: 10.1109/ SURV.2012.110112.00192 .

134] R. Wright , L. Keith , Wearable technology: if the tech fits, wear it, J. Electron.

Res. Med. Lib. 11 (4) (2014) 204–216 . 135] W. Feng, Z. Yan, H. Zhang, K. Zeng, Y. Xiao, Y.T. Hou, A survey on security, pri-

vacy, and trust in mobile crowdsourcing, IEEE Internet Things J. 5 (4) (2018) 2971–2992, doi: 10.1109/JIOT.2017.2765699 .

136] P. Datta, A.S. Namin, M. Chatterjee, A survey of privacy concerns in wearable devices, in: 2018 IEEE International Conference on Big Data (Big Data), 2018,

pp. 4549–4553, doi: 10.1109/BigData.2018.8622110 . 10-13 Dec. 2018 .

[137] S. Jhajharia , S. Pal , S. Verma , Wearable computing and its application, Int. J. Comput. Sci. Inf. Technol. 5 (4) (2014) 5700–5704 .

138] “http://pop.h- cdn.co/assets/cm/15/13/5512f7e752a4e _ - _ Wearables- and- the- Science- of- Human- Behavior- Change- EP4.pdf ,” Accessed

July 12, 2021. 139] S. Kekade , et al. , The usefulness and actual use of wearable devices among the

elderly population, Comput. Methods Programs Biomed. 153 (2018) 137–159 .

140] J.H.M. Bergmann, A.H. McGregor, Body-worn sensor design: what do patients and clinicians want? Ann. Biomed. Eng. 39 (9) (2011) 2299–2312, doi: 10.1007/

s10439- 011- 0339- 9 . [141] E.J. Corwin, D.P. Jones, A.L. Dunlop, Symptom science research in the era of

big data: leveraging interdisciplinary resources and partners to make it hap- pen, J. Nurs. Scholarsh. 51 (1) (2019) 4–8, doi: 10.1111/jnu.12446 .

142] M. Hassanalieragh, et al., Health monitoring and management using internet-

of-things (IoT) sensing with cloud-based processing: opportunities and chal- lenges, in: 2015 IEEE International Conference on Services Computing, 2015,

pp. 285–292, doi: 10.1109/SCC.2015.47 . 27 June-2 July 2015 . 143] Z. Lv, J. Chirivella, P. Gagliardo, Bigdata oriented multimedia mobile health

applications, J. Med. Syst. 40 (5) (2016) 120, doi: 10.1007/s10916- 016- 0475- 8 . 144] Building the hospital without walls - a CSIRO home telecare initiative,

Telemed. J. 6 (2) (20 0 0) 275–281, doi: 10.1089/107830200415225 .

145] I. Kiral-Kornek, et al., Epileptic seizure prediction using big data and deep learning: toward a mobile system, EBioMedicine 27 (2018) 103–111, doi: 10.

1016/j.ebiom.2017.11.032 .

https://doi.org/10.1109/TITB.2009.2036164

https://doi.org/10.1109/JBHI.2014.2307584

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0113

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0114

https://doi.org/10.1016/j.sigpro.2017.10.011

https://doi.org/10.1007/s10845-018-1416-9

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0117

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0118

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0119

https://doi.org/10.1007/978-3-540-74048-3_4

http://refhub.elsevier.com/S0169-2607(21)00615-5/sbref0121