See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/328897156

Artificial Intelligence for Long-term Respiratory Disease Management

Conference Paper · July 2018

DOI: 10.14236/ewic/HCI2018.65

CITATIONS READS

3 1,305

3 authors:

Philip Catherwood Joseph Rafferty

Ulster University Ulster University
33 PUBLICATIONS   334 CITATIONS    54 PUBLICATIONS   448 CITATIONS   

SEE PROFILE SEE PROFILE

James Andrew McLaughlin

Ulster University
390 PUBLICATIONS   7,939 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Sensors: Special Issue "Smart Homes: A Prospective of Sensing, Communication, and Automation" View project

Potential for subcutaneous implants View project

All content following this page was uploaded by Philip Catherwood on 04 March 2019.

The user has requested enhancement of the downloaded file.

 http://dx.doi.org/10.14236/ewic/HCI2018.65

Artificial Intelligence for Long-term Respiratory

Disease Management

Philip A. Catherwood Joseph Rafferty James McLaughlin

School of Engineering Sch. of Computing & Mathematics School of Engineering
Ulster University Ulster University Ulster University
Shore Rd., N’abbey Shore Rd., N’abbey Shore Rd., N’abbey
Co.Antrim, BT370QB Co.Antrim, BT370QB Co.Antrim, BT370QB
[email protected] [email protected] [email protected]

This paper presents the strengths, weaknesses, opportunities, and threats for Artificial Intelligence
(AI) as applied to long-term respiratory disease management. This analysis will help to identify,
understand, and evaluate key aspects of the technology as well as the various internal/external
forces which influence its success in this application space. Such understanding is instrumental to
ensure judicial planning and implementation with suitable safeguards being considered. AI has the
potential to radically change how respiratory disease management is conducted and may help
clinicians to realise new treatment paradigms. The application of AI is clearly not specific to
respiratory disease management; however it is a chronic disease that requires on-going
monitoring and well evidenced decision making regarding treatment pathways or medication
modification. This work emphasises the current position of AI as applied to respiratory disease
management and identifies the issues to help develop strategic directions to ensure successful
implementation, evidenced by ubiquitous acceptance and uptake.

Artificial Intelligence, Disease management, Machine learning, Respiratory, SWOT

1. INTRODUCTION The subject of this recent research has focused

around machine learning-based approaches to
As medical technology gets smaller, smarter, and classification and recognition of activities, notably
more able to process large volumes of data, there within the area of Big Data. Generally, these
are opportunities to take devices which are machine learning approaches are within 2 broad
hospital-based and realise them as a small portable categories - supervised and unsupervised.
devices, facilitating mobile and remote healthcare Supervised machine learning builds a classification
strategies. Additionally there is a growing trend to model through providing it with annotated data,
embed intelligence into such devices enabling them such as images tagged with enumeration
to make decisions based on ongoing preferences representation conditions, such as healthy skin,
and inputs, such as biomedical measurements. To unhealthy skin, or indeterminate. This model is then
this end a software platform, not a clinician, may applied to previously unseen data to provide a
inform or produce medical decisions about classification.
treatment and interventions. Furthermore,
emergence of pervasive communications and Unsupervised machine learning leverages
Internet of Things (IoT) technologies ensure unlabelled data to correlate data within a dataset.
significant data can be relayed to clinical experts This enables analysis of data such as matching
and cloud-based software platforms enabling similar data in groups/clusters or identification of
decision support. anomalies/outliers. Yu et al. [1] leveraged
supervised machine learning to produce a model
There has been a wide range of recent studies which accurately classified the potential for
where AI has been applied to assist with clinical dermoscopy images to be cancerous, Deep
decisions such as the automated diagnosis of skin Convolutional Neural Networks (CNN) were used to
cancer [1], bone abnormalities [2], assistance with realise this solution producing a viable trained
DNA sequencing [3], and supporting surgery model from 900 images. Jagadev et al. [2]
scheduling [4]. Diagnosis of obstructive lung implemented supervised machine learning to
disease is an area of potential promise [5]; it could produce a Support Vector Machine-based classifier
increase detection accuracy [6] and automate capable of identifying images of blood smear tests
screening to increase patient throughput. which are likely to have a form of leukaemia.

© Catherwood et al. Published by

BCS Learning and Development Ltd. 1
Proceedings of British HCI 2018. Belfast, UK.
 Artificial Intelligence for Long-term Respiratory Disease Management
Catherwood ● Rafferty ● McLaughlin

Celesti et al. [3] surveyed studies which involved Given the types of data and availability of datasets
AI-based analysis of genomics data to identify a swathe of AI techniques can be used within this
conditions and anomalies. These surveyed studies domain. Supervised machine learning may be used
used supervised and unsupervised machine to build a classification model which may be used
learning as dictated by their use case. Typically, to indicate when COPD symptoms are likely to
supervised learning was employed to identify deteriorate. This model may be produced from an
conditions using a model built from training data annotated dataset leveraging environmental
whereas unsupervised machine learning was used metrics, biomarkers, an individual’s medical history
to identify anomalies. and current state. Unsupervised machine learning
may be applied to a large dataset gathered from
In recent years AI research has had a focus on
sufferers to help identify factors which can
using machine learning, however, works exist
contribute to, or prevent, COPD flare ups.
which have devised Decision Support Systems
(DSS) support systems and traditional, tailored, A DSS may be produced from domain knowledge
classification algorithms. Xiang et al. [4] have to inform, prompt, and advise respiratory disease
produced a DSS incorporating custom classification suffers in a tailored fashion. This DSS may take
algorithms to optimise scheduling of surgical inputs from PEF measurements, environmental
procedures giving operational restraints and data, and supervised machine learning
required resources, such as availability of an classification of an individual’s parameters. The
operation theatre and supporting staff. This has realisation of such a solution will be the subject of
proven to optimise efficiency and increase health future work and will be informed by the quality and
outcomes. quantity of data that can be reliably obtained from
sufferers of respiratory diseases.
All illness management may benefit from AI;
respiratory disease management is an enduring
key medical theme which encompasses Asthma, 2. S.W.O.T. ANALYSIS
chronic obstructive pulmonary disease (COPD),
pulmonary hypertension, bronchiectasis, sleep An analysis of the strengths, weaknesses,
apnoea, etc. [7] and attributes to around 20% of UK opportunities, and threats of AI’s application to
deaths [8]. COPD is ranked as the third most respiratory disease management is an essential
common global cause of death in the last decade aspect of understanding the potential of the
[9]. technology as well as the paths to application. The
strengths and weaknesses are with respect to the
Respiratory disease is commonly diagnosed and
monitored using Spirometry (lung function test) internal aspects of AI. Any identified weaknesses
which evaluates the volume of air expelled from the may be addressed via technological revisions. The
opportunities and threats are with respect to the
lungs (forced vital capacity) as well as the rate at
which the air is expelled (forced expiratory volume) external aspects; these include other technologies,
[10]. A similar, simpler, meter measuring the peak public perception of the technology, financial
climate, etc. Threats are addressed via methods
expiratory flow (PEF) is often used in everyday
clinical practice (often referred to as peak flow such as legislation and public education strategies,
meter). This technology has been mechanical in although are more difficult to change. The use of
SWOT analysis gives a panoramic and
nature but in recent years has been realised as
electronic measurement devices with wireless comprehensive perspective of the state of the
connectivity, typically Bluetooth. Other diagnosis technology in the specific sector. It is a universal
techniques which complement spirometry include method which creates a usable examination which
the use of biomarkers for COPD or lung cancer [11, may act as a starting point to tackle the challenges.
12]. Such biomarkers may be realised as
2.1 Strengths
implantable monitors or home test kits; employing
AI would make a powerful point-of-care system. AI boasts the ability to recognise objects in images,
There are clear correlations between air pollution transcribe speech, control machines, analyse DNA
and respiratory disease, mortalities, and outpatient to detect genomic conditions, etc. [17]. Such
admissions [13]. Approximately 3 million people die technological strengths expedite the delivery of
each year from COPD, which is estimated to be bespoke personalised care for respiratory disease
around 6% of all deaths worldwide. It is also management. AI’s ability to learn, find patterns,
determined that some 235 million people suffer make decisions, etc. This may be facilitated
from asthma [7]. While significant progress has through sensor measurements from spirometers,
been made to predict, monitor and treat respiratory inhaler and medication usage, biomarkers, etc. In
disease [14-16] there is still significant exploration addition, qualitative data may be incorporated from
required. active assistants/chat bots to understand feelings,
pain, anxiety, breathlessness, etc. Such solutions
will assist with making informed machine decisions

2
 Artificial Intelligence for Long-term Respiratory Disease Management
Catherwood ● Rafferty ● McLaughlin

leading to medication alteration, clinician keep patients out of hospital [18]. Smart homes are
intervention or hospitalisation. AI could predict a increasing popular [23] and there may be
respiratory disease decline, such as an asthma opportunities to integrate medical devices with
attack, before the patient even considers there to these networks. Implementing smart healthcare
be anything untoward or peak flow measurements devices into the community also helps to drive
show a notable decline in lung function. The technology and services into remote areas to
abundant recorded data that can be stored from the address the recognised issues of rural communities
vast array of patients can help the training of being last to get the latest technical services [24].
accurate AI models, particularly through deep Having intelligent sensors and systems that are
learning approaches. It is recognised that early continually monitoring your wellness is likely to
detection of exacerbations in COPD can increase have measurable health benefits over occasional
positive outcomes and reduce hospital admissions; clinic visits; such monitoring resolution will assist
telehealth-based systems interventions can with faster responses to deteriorations as well as
decrease the costs associated with COPD patients better diagnoses. This principle has been proven
[18] and promote better self-management [19, 20]. already with smart implanted insulin pumps which
actively monitor blood glucose levels and
2.2 Weaknesses automatically medicate to exacting levels
accordingly [25]. Further emerging technologies
In late 2017 mainstream technical news sources such as stretchable electronics [26] and long range
published reports with taglines such as “No one low power communication systems [27] increase
really knows how the most advanced algorithms do the potential for effective technology deployment.
what they do” [21]. This suggests that the decisions
that AI makes can be unexplainable or beyond 2.4 Threats
logic. For example, chip maker Nvidia tested a car
controlled by AI and later reported that the vehicle The literature shows limited work on AI and
didn’t follow a single instruction provided by the telemetry for respiratory disease monitoring [28]
programmers, instead using an algorithm that it had which means it is not as well understood as some
taught itself. While this may be good for attention- other areas of study. However, some key
grabbing headlines it creates disquiet for those who explorations on the use of machine learning for
are looking to AI to intelligently manage the health respiratory disease diagnosis in clinical settings
decisions of the chronically ill. The vehicle in the serve to effectively highlight the value of developing
test conducted itself appropriately and completed the technology [29-31]. Other external threats may
the task very well but the engineers couldn’t impede the uptake of AI for respiratory disease
describe how it did it [21]. This is not an isolated management including the fear of lawsuits against
report but a growing trend [22]. Furthermore, some the system creator if AI is proven to have made a
complex AI systems have been described as poor judgement or an obvious mistake. Likewise,
“inscrutable” as it is not possible to make the AI the over-expectation of investors, technical
system explain why it made certain decisions; such communities and medical professionals, fears over
inability for humans to understand the decision- security of data and hacking, poor investment into
making processes of the system creates significant the technology, user error of any aspect of the
risk. These inscrutable solutions typically leverage system, issues with poor battery performance for
machine learning based approaches to AI. When portable devices, and global economics which
that risk is applied to chronic diseases like COPD demands continual cost cutting will all serve to
such inscrutable systems are unacceptable and it threaten successful adoption of the technology.
would not be possible to verify if a death was an High profile figures such as the late Professor
unavoidable outcome or a bizarre AI decision. Stephen Hawking have done little to help the public
Validation of medical technology requires perception of AI [32]. Furthermore the notoriously
substantial clinical testing which costs time and slow uptake of new technology in healthcare [33]
money; however system reliability and observability may result in the technology being quickly
are two of the key parameters of any system under outdated, especially in the context of AI technology
test. AI in healthcare cannot be a leap of faith into changing so rapidly.
the unknown and capricious systems may have no
A notable high profile AI failure will no doubt cast a
future in this space. To this end, such solutions
long shadow over the trust of AI implementation for
may be limited to offer insight to inform decisions to
some time; in March 2018 an autonomous car
be made by medical practitioners.
being driven by AI was involved in a fatal
2.3 Opportunities pedestrian accident [34]. Internet commentators
reasoned that the outcome would have been the
With respiratory disease being a significant strain same regardless of whether a computer or a
on healthcare resources it is advantageous to push person was under control of the vehicle,
the monitoring of patients into the community to nonetheless the headline will still capture the
imagination of those already fearful of machines

3
 Artificial Intelligence for Long-term Respiratory Disease Management
Catherwood ● Rafferty ● McLaughlin

making life and death decisions. These issues are [4] Xiang, W., Yin, J., Lim, G. (2015) A short-term
real and cannot be overcome easily; objections will operating room surgery scheduling problem
remain and as such the technology could struggle integrating multiple nurses roster constraints. Art.
to enjoy widespread acceptance without significant Int. in Med., vol. 63, pp. 91-106.
research and validation of medical AI systems.
While automotive AI applications are clearly distinct [5] Das, N., Topalovic, M., Janssens, W. (2018)
from healthcare applications the public perception Artificial intelligence in diagnosis of obstructive lung
may not be adequately informed to appreciate the disease: current status and future potential. Curr.
different techniques and technologies utilised. Opin. Pulm. Med. vol. 24, pp. 117-123.

[6] IEEE Spectrum News. (2017) AI versus doctors.

3. CONCLUSIONS spectrum.ieee.org/static/ai-vs-doctors (6/3/18).
[7] British Lung Foundation (2018) Lung disease in
The potential for impact and reduced costs for the UK. https://statistics.blf.org.uk/lung-disease-uk-
respiratory disease management is clear, however big-picture (5/6/18)
there are also significant obstacles to
implementation and the approval processes in [8] WHO. (2018) Chronic respiratory diseases.
healthcare could make the new technology lag http://www.who.int/respiratory/en/ (6/3/18).
behind AI implementation in other sectors;
healthcare could therefore be the last sector to [9] Patel, J., et al. (2014) Global and regional
experience the benefits of the IoT/AI revolution. trends in mortality from chronic obstructive
Correct implementation however will fundamentally pulmonary disease: their relation to poverty,
enable fast reactions to emergencies in the short smoking and population change. Eur. Respir.
term and disease progression management in the Journal. vol. 44, pp. 1239–1247.
long term. It will also have the capacity to address
the increased management complications [10] Komarow, H.D., et al. (2011) Impulse
presented by multiple chronic conditions (MCC). oscillometry in the evaluation of diseases of the
Non-medical devices purporting to contain AI may airways in children. Ann. Allergy Asthma Immunol.,
serve to impede authorised application in clinical vol. 106, pp. 191-199.
practice as if these consumer electronics lifestyle
devices are of questionable trustworthiness they [11] Khan, K., et al. (2017) Potential biomarkers for
may complicate acceptance of AI in automated the diagnosis of respiratory tract infection and lungs
healthcare decision making. General cancer. Cell Molecular Biology. vol. 30, pp. 46-52.
recommendations for well-managed
implementation includes ensuring the technology is [12] Kim, H.J., Choi, M.G., Park, M.K., Seo, Y.S.
not hurried to meet industry expectations, the use (2017) Predictive and Prognostic Biomarkers of
of valid training datasets taken from a cross section Respiratory Diseases due to Particulate Matter
of communities and diseases, ongoing healthcare Exposure. J. Cancer Prev. vol. 22, pp. 6–15.
cost evaluations, and the acknowledgement that AI
is still in its infancy and has significant maturing to [13] Mo, Z., et al. (2018) Acute effects of air
do before being ready for such an audacious task. pollution on respiratory disease mortalities and
outpatients in Southeastern China. Nature. vol. 8,
pp. 1-9. Springer, USA.
4. REFERENCES
[14] Giannaccini, M.E., et al. (2017) Respiratory
[1] Yu, L., Chen, H., Dou, Q., Qin J., Heng, P.A. simulator for robotic respiratory tract treatments.
(2017) Automated Melanoma Recognition in IEEE Intl. Conf. Robotics and Biomimetics, Macau,
Dermoscopy Images via Very Deep Residual Macao, 5-8 Dec. 2017, pp. 2314-2319, IEEE, USA..
Networks. IEEE T. Med. Im., vol. 36, pp. 994-1004.
[15] Rehouma, H., Noumeir, R., Jouvet, P.,
[2] Jagadev P., Virani, H.G. (2017) Detection of Bouachir, W., Essouri, S. (2017) A computer vision
leukemia and its types using image processing and method for respiratory monitoring in intensive care
machine learning, Intl. Conf. Trends Elect. Info. environment using RGB-D cameras. 7th Intl. Conf.
(ICEI), Tirunelveli, pp. 522-526. IEEE, USA. Im. Proc. Theory, Tools and App., Montreal, QC,
28th Nov.-1st Dec. 2017, pp. 1-6, IEEE, USA.
[3] Celesti, F., et al., (2017) Big data analytics in
genomics: The point on Deep Learning solutions. [16] Nousias S., et al., (2018) An mHealth System
IEEE Symp. Comp. Comms. (ISCC), Heraklion, pp. for Monitoring Medication Adherence in Obstructive
306-309. IEEE, USA. Respiratory Diseases Using Content Based Audio
Classification. IEEE Access, vol. 6, pp. 11871-
11882.

4
 Artificial Intelligence for Long-term Respiratory Disease Management
Catherwood ● Rafferty ● McLaughlin

[17] Newton-Rex, E. (2017) Business Insider. [29] Prasad B., Prasad P.K., Sagar, Y. (2011)
http://uk.businessinsider.com/artificial-intelligence-ai- Advances in computer science and information
most-impressive-achievements-2017- technology. Springer, Berlin, pp. 570–576.
3?r=US&IR=T/#what-ai-cando-everyday-humanstuff-1
(24/3/18). [30] Pescatore A.M., et al. (2014) A simple asthma
prediction tool for preschool children with wheeze
[18] Guarascio, A.J., Ray, S.M., Finch, C.K. (2013) or cough. J All. Clin. Immun. vol. 133, pp. 111–118.
The clinical and economic burden of chronic
obstructive pulmonary disease in the USA. [31] Badnjevic A, et al. (2015) Neuro-fuzzy
Clinicoecon Outcomes Res. vol. 5, pp. 235–245. classification of asthma and chronic obstructive
pulmonary disease. BMC Med. Inform. Decision
[19] Fernandez-Granero, M.A., Sanchez-Morillo D., Making. vol. 15, pp. S1.
Leon-Jimenez A. (2018) An artificial intelligence
approach to early predict symptom-based [32] Cellan-Jones, R. (2014) BBC.
exacerbations of COPD. Biotech. & Biotech. www.bbc.co.uk/news/technology-30290540 (19/2/18)
Equipment, DOI: 10.1080/13102818.2018.1437568
[33] Palabindala, V., Pamarthy, A., Jonnalagadda,
[20] Esteban, C., et al. (2015) Machine learning for N.R. (2016) Adoption of electronic health records
COPD exacerbation prediction. Eur. Resp. J. vol. and barriers. J. Community Hosp. Intern. Med.
46: OA3282. Perspect. vol. 6, DOI: 10.3402/jchimp.v6.32643

[21] Knight, W. (2017) MIT. [34] Wakabayashimarch, D. (2018) NY Times.

www.technologyreview.com/s/604087/the-dark- https://www.nytimes.com/2018/03/19/technology/uber
secret-at-the-heart-of-ai/ (3/3/18) -driverless-fatality.html (29/3/18).

[22] Gershgorn, D. (2017) Quartz,

qz.com/1146753/ai-is-now-so-complex-its-creators-
cant-trust-why-it-makes-decisions/ (3/3/18)

[23] Wetzel, K. (2018) Digital Trends

www.digitaltrends.com/home/smart-home-technology-
trends-and-whats-ahead/ (21/3/18).

[24] Ofcom(2017) Ofcom www.ofcom.org.uk/about-

ofcom/latest/media/media-releases/2017/northern-
ireland-broadband (21/3/18)

[25] Pickup, J., Holloway, M., Samsi, K. (2015)

Real-time continuous glucose monitoring in type 1
diabetes: a qualitative framework analysis of
patient narratives. Dia. Care. vol. 38, pp. 544–550.

[26] Son, D., et al, (2014) Multifunctional

wearable devices for diagnosis and therapy of
movement disorders. Nature Nanotech. vol. 9, pp.
397-404.

[27] Catherwood, P., McComb, S., Little, M.,

McLaughlin, J, (2017). Channel Characterisation
for Wearable LoRaWAN Monitors, Loughborough
Ant. and Prop. Conf. (LAPC), 12-13 Nov. 2017, pp.
1-4. IET, UK.

[28] Sanchez-Morillo, D., Fernandez-Granero,

M.A., Leon-Jimenez, A. (2016) Use of predictive
algorithms in-home monitoring of chronic
obstructive pulmonary disease and asthma.
Chronic Resp. Disease. vol. 13, pp. 264-283.

View publication stats