Artificial Intelligence in Medicine

Pavel Hamet, Johanne Tremblay

PII: S0026-0495(17)30015-X
DOI: doi: 10.1016/j.metabol.2017.01.011
Reference: YMETA 53528

To appear in: Metabolism

Please cite this article as: Hamet Pavel, Tremblay Johanne, Artificial Intelligence in
Medicine, Metabolism (2017), doi: 10.1016/j.metabol.2017.01.011

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Artificial Intelligence in Medicine

Pavel Hamet OQ, MD, PhD and Johanne Tremblay PhD

T
Authors affiliations : Centre de recherche, Centre hospitalier de l’Université de Montréal

P
(CRCHUM), Montréal, Québec, Canada, H2X 0A9 and Department of Medicine, Université de

RI
Montréal, Montréal, Québec, Canada, H3T 3J7

SC
NU
Correspondence to: Dr Pavel Hamet, CRCHUM room R14.404 or Dr Johanne Tremblay,
MA
CRCHUM, room R8.456, 900 Saint-Denis Street, Montréal, Québec, Canada H2X 0A9.

Telephone +1 514-890-8246,
ED

www.crchum.com
PT

pavel.hamet@umontreal.ca
CE

johanne.tremblay@umontreal.ca
AC

Words: 2560
 ACCEPTED MANUSCRIPT

Abstract

Artificial Intelligence (AI) is a general term that implies the use of a computer to model

intelligent behaviour with minimal human intervention. AI is generally accepted as having started

P T
with the invention of robots. The term derives from the Czech word robota, meaning biosynthetic

RI
machines used as forced labour. In this field, Leonardo Da Vinci’s lasting heritage is today’s

SC
burgeoning use of robotic-assisted surgery, named after him, for complex urologic and

gynecologic procedures. Da Vinci’s sketchbooks of robots helped set the stage for this

NU
innovation. AI, described as the science and engineering of making intelligent machines, was
MA
officially born in 1956. The term is applicable to a broad range of items in medicine such as

robotics, medical diagnosis, medical statistics, and human biology—up to and including today’s
ED

“omics”. AI in medicine, which is the focus of this review, has two main branches: virtual and

physical. The virtual branch includes informatics approaches from deep learning information
PT

management to control of health management systems, including electronic health records, and
CE

active guidance of physicians in their treatment decisions. The physical branch is best

represented by robots used to assist the elderly patient or the attending surgeon. Also embodied in
AC

this branch are targeted nanorobots, a unique new drug delivery system. The societal and ethical

complexities of these applications require further reflection, proof of their medical utility,

economic value, and development of interdisciplinary strategies for their wider application.

Keywords: artificial intelligence, robots, future of medicine, avatars

2
 ACCEPTED MANUSCRIPT

Introduction

Artificial intelligence (AI) is generally accepted as having started with the invention of robots.

The word robot, spelled robota in Czech, was introduced into the literature by the writer Karel

P T
Capek in his 1921 play, “R.U.R” (Rossum’s Universal Robots). It signified a factory where

RI
biosynthetic machines are used as forced labor. In the middle of the last century, Isaac Asimov

SC
immortalized the word “robot” in a collection of short stories of modern science-fiction. The first

mention of a humanoid automaton, however, can be traced back to the third century in China

NU
when a mechanical engineer, Yan Shi, presented to the Emperor Mu of Zhou, a human shaped
MA
figure of mechanical handiwork built with leather, wood and artificial organs. [1] In the 12th

century, a Muslim golden age scholar, polymath, inventor, and mechanical engineer named al-
ED

Jazari created a humanoid robot able to strike cymbals. During the Renaissance period, Leonardo

da Vinci made a detailed study of human anatomy to design his humanoid robot. His sketches
PT

drawn in 1495, were only rediscovered in the 1950s. Leonardo’s robot was a knight robot that
CE

was able to stand-up, sit-down, wave arms and move head and jaw. It was operated by pulleys

and cables. More important than his accomplishments in this area, da Vinci’s sketchbooks were a
AC

source of inspiration for a generation of robotic researchers, some of whom worked at NASA.

In medicine, a surgical system made by the American company, Intuitive Surgical, was named

Da Vinci in recognition of his inspirational impact. It was approved by the Food and Drug

Administration (FDA) in 2000, and the number of units in operation around the world is now

over 5000. Da Vinci surgical systems facilitate complex surgery using a minimally

invasive approach, and can be controlled by a surgeon from a console. The system is commonly

used for prostatectomies and gynecologic surgical procedures. It is starting to be used for cardiac

valve repair.

3
 ACCEPTED MANUSCRIPT

The evolution of robots made a change in direction with the first robot to be recognized as

revolutionary in its mechanical realistic conception being the “Flute Player”, conceived in the

18th century by the French inventor, Jacques de Vaucanson, as an innovative “automaton”

T
playing the pipe. It had a repertoire of 12 songs. Two centuries later, William Grey Water

P
RI
invented the first electronic autonomous robots. His goal was to demonstrate how the brain

functions. He became famous in 1948 for the fabrication of the first electronic autonomous

SC
robot, that he named Machina Speculatrix. It revealed that connections between a small number

NU
of “brain cells” could lead to very complex behaviors. John McCarthy coined the term "artificial

intelligence" (AI) in 1955, defining it as "the science and engineering of making intelligent
MA
machines". He was very influential in the early development of AI. With his colleagues he

founded the field of AI in 1956 at a Dartmouth College conference on artificial intelligence. The
ED

conference gave birth to what developed into a new interdisciplinary research area. It provided an
PT

intellectual framework for all subsequent computer research and development efforts.
CE

During the following years, computers started to solve many complex mathematical

problems that soon became of interest to the Department of Defense of the USA. Then, after a
AC

period of slowdowns in the 80’s, a new golden era restarted with the use of logistic data mining

and medical diagnosis. Instruments with increasing computational power were developed. This

new capability allowed Big Blue to finally beat the world Chess champion, Gary Kasparov on

May 11, 1997.

Today, AI is considered a branch of engineering that implements novel concepts and novel

solutions to resolve complex challenges. With continued progress in electronic speed, capacity,

and software programming, computers might someday be as intelligent as humans. One cannot

neglect the important contribution of contemporary cybernetics to the development of AI.

4
 ACCEPTED MANUSCRIPT

Defined as a trans-disciplinary approach, cybernetics aims for control of any system using

technology that explores system regulation, structure and constraints, most notably mechanical,

physical, biological, and social. The origin of cybernetics is attributed to Norbert Wiener [2], who

T
formalized the notion of feedback, with implications for engineering, systems control, computer

P
RI
science, biology, neuroscience, philosophy, and the organization of society. Fields that were most

influenced by cybernetics are (if we exclude game theory) systems theory, sociology, psychology

SC
(especially neuropsychology and cognitive psychology), and theory of organizations.

NU
Today literature on AI is abundant and unbridled. AI was portrayed as a possible threat to the
MA
world economy during the 2015 economic forum held at Davos, where Stephen Hawking even

expressed his fear that AI may one day eliminate humanity [3]. We will not discuss here the use
ED

of this rapidly developing field in military, security, transport or manufacturing, instead, the

focus of our chapter is on medicine and health systems.

PT

Artificial intelligence in medicine: The virtual branch

CE

The application of AI in medicine has two main branches: virtual and physical. The virtual
AC

component is represented by Machine Learning, (also called Deep Learning) that is represented

by mathematical algorithms that improve learning through experience. There are three types of

machine learning algorithms: (i) unsupervised (ability to find patterns), (ii) supervised

(classification and prediction algorithms based on previous examples), and (iii) reinforcement

learning (use of sequences of rewards and punishments to form a strategy for operation in a

specific problem space). First, AI has boosted and is still boosting discoveries in genetics and

molecular medicine by providing machine learning algorithms and knowledge management. An

example of successes in medicine is the unsupervised protein-protein interaction algorithms that

led to novel therapeutic target discoveries [4]. The methodology used a combination of adaptive
5
 ACCEPTED MANUSCRIPT

evolutionary algorithms and state-of-the-art clustering methods, named “evolutionary enhanced

Markov clustering”. It permitted prediction of over 5000 protein complexes, of which over 70%

were enriched by at least one gene ontology function term [4]. Novel computational methodology

T
is also being developed to identify DNA variants such as single nucleotide polymorphisms

P
RI
(SNPs) as predictors of diseases or traits, using novel evolutionary embedded algorithms that are

more robust and less prone to over-fitting issues that occur when a model has too many

SC
parameters relative to the number of observations [5].

NU
Today’s “systems thinking” about health care focuses not only on the classical interactions
MA
between patients and providers but takes into account larger-scale organizations and cycles.

Furthermore, the health care system must not be stationary but must learn from its own
ED

experiences and strive to implement continuous process improvements. This is a multi-agent

system (MAS), where a set of agents situated in a common environment interact with each other.
PT

This process involves building or participating in an organization, which uses AI to achieve

CE

significant progress.

An example of such a process in medicine is the development of problematically complex

AC

ecosystems for treating chronic mental diseases [6]. Instead of focusing on health expenditures

(in public health systems) or cost recovery (in health management organization), the MAS

approach proposes to capture the dynamics of individual patients, including their responses to

received medications as well as their behavioral interactions within a larger societal ecosystem.

This global care coordination technology allows process mapping, facilitates control, and better

supports changes to the system with a demonstrated increase in response to medication, decrease

of costs and more efficacious interventions. Its implementation has allowed health systems

6
 ACCEPTED MANUSCRIPT

managers to analyse the dynamics of system performance across changes in social, medical and

criminal justice components [7].

Included in the virtual applications of AI are electronic medical records where specific

P T
algorithms are used to identify subjects with a family history of a hereditary disease or an

RI
augmented risk of a chronic disease. AI is used to improve organizational performance by

SC
enabling individuals to capture, share and apply their collective knowledge to make “optimal

decisions in real time”. As a consequence, electronic medical records and health care process

NU
management are crucial to achieve the desired quality. From current patients’ record keeping of
MA
variable quality, information needs to be captured in a digital format accessible as individual data

as well as in aggregated forms for epidemiological research and planning. Major efforts are
ED

required from academia and the information technology industry to achieve desired efficacy and

minimize cost.
PT

The current status of medical records is mostly in the form of incommunicable silos of wasted
CE

information for the health system and for knowledge acquisition. Laboratories and clinics need

to collaborate to accelerate the implementation of electronic health records [8]. Data need to be
AC

captured in real-time, and institutions should promote their transformation into intelligible

processes. New scientific and clinical findings should be shared through open-source, and

aggregated data must be displayed for open-access by physicians and scientists and made

automatically available as point-of-care information. Integration and interoperability including

ethical, legal and logistical concerns are enormous, particularly with the forthcoming addition of

“omics-based” data. The simplification, readability and clinical utility of data sets should be

made evident, and each result must be questioned for its clinical applicability. Our group has

developed such a design as illustrated in Figure 1. Electronic medical or health records are

7
 ACCEPTED MANUSCRIPT

essential tools for personalized medicine and for early detection and targeted prevention, again

with the aim of increasing their clinical value and decreasing health costs [9].

Further virtual application of AI in medicine is the use of softbots, as psychotherapeutic

P T
avatars [10]. Avatars stem from the famous 2009 James Cameron movie which features a hybrid

RI
human-alien created to facilitate communication with people from the planet known as Pandora.

SC
The use of emotionally sensitive teachable avatars is receiving recognition in medicine [11]. It

has been applied to pain control in children with cancer (called “pain body”) and it is able to

NU
detect early emotional disturbances in youngsters in native American reservations, including
MA
suicidal tendency. This approach seems to work better than human interventions [11]. One of the

clearest examples is the control of paranoid hallucinations when the subject designs his own
ED

avatar representing the persecutor in his mind. The system encourages the subject to engage in

discussions with his persecutor who progressively learns to moderate such destructive behavior.
PT

Initial successes with this technology have been demonstrated by achievement of a lower level of
CE

hallucinations and vocal threats. Perhaps the most useful function will be in care of the elderly,

where the frequency, reassuring nature, and kindness of what is said are all important elements of
AC

improved communication. Avatars have been also applied to home care, and for biological and

physical monitoring with 3D vision [12].

Artificial intelligence in medicine: The physical branch

The second form of application of AI in medicine includes physical objects, medical devices

and increasingly sophisticated robots taking part in the delivery of care (carebots) [13]. Perhaps

the most promising approach is the use of robots as helpers; for example, a robot companion for

the aging population with cognitive decline or limited mobility. Japanese carebots are the most

advanced forms of this technology. Robots are used in surgery as assistant-surgeons or even as
8
 ACCEPTED MANUSCRIPT

solo performers [14]. One of the most impressive examples of the utility of robots is their ability

to communicate with and teach autistic children. Here, and in many other situations that might

benefit from robotic intervention, important ethical considerations will have to be resolved before

T
it will become possible to use AI-robots routinely in today’s medical environment. Apart from

P
RI
ethical issues, a major challenge in this new dimension of medical care is the clear need for

standardized, comparative evaluation of the effect of robotic systems on health indicators, and

SC
measures of changes in psychological and physical status, side effects, and outcomes [15].

NU
Use of robots to monitor effectiveness of treatment
MA
Robots can also be useful in the evaluation of changes in human performance in such

situations as rehabilitation [16]. Another area where AI might be helpfully employed is for
ED

monitoring the guided delivery of drugs to target organs, tissues or tumours. For example, it is

encouraging to learn of the recent development of nanorobots designed to overcome delivery

PT

problems that arise when difficulty of diffusion of the therapeutic agent into a site of interest is
CE

encountered. This problem occurs when the therapist is attempting to target the core of a tumour

which tends to be less vascularized, anoxic, but most proliferatively active. To overcome
AC

limitations of mechanical or radioactive robotics, researchers have attempted to harness a natural

agent with desired properties as a replacement of “intelligent” nanoparticles alone. For this

purpose, they are studying a special type of marine coli, called Magnetococcus marinus which

travels spontaneously to low oxygenated zones. Initial guidance is provided by an external

magnetic source and then inherent properties of nanorobots are put into play. These nanorobots

can be covalently bound with nanoliposomes bearing therapeutic properties. Early data have

disclosed a significant increase in the gradient of desired drug into the hypoxic zones [17].

9
 ACCEPTED MANUSCRIPT

Most of these novel applications of AI in medicine need further research, particularly in areas

of human-computer interactions. Moshimo Mori [18] introduced in 1970 the notion of uncanny

valley in which an important theme is the human-robot interaction (HRI) field. In these studies

T
humanoid robots were evaluated for their apparent humanity, eeriness and attractiveness as

P
RI
factors making perception of robots either acceptable, feared or rejected [19].

SC
Conclusion

NU
AI for personal use is going to stay with us much as genetics will continue to provide

personal services. It is therefore important to consider how AI will also serve the development of
MA
our health care systems. Takashi Kido [20] proposed MyFinder as a personalized community

computing to resolve challenges of personalized genome services, acting jointly with AI and
ED

shaping the personalized and participative health care of the future. The goal of this platform is

to provide personal genome environment interaction in both directions: impact of genes on

PT

diseases, health and drug responses, and impact of our environment, behavioral and wellness on
CE

our gene activities. The World Economic 2016 Forum named open AI ecosystem as one of the 10

most important emerging technologies. With the unprecedented amount of data available,
AC

combined with advances in natural language processing and social awareness algorithms,

applications of AI will become increasingly more useful to consumers. This is particularly true in

medicine and healthcare where there are many data to be utilized from patient medical records

and lately also from information obtained by wearable health sensors. This huge volume of data

should be analyzed in detail, not only to provide patients who want suggestions about lifestyle,

but also to generate information aimed at improving healthcare design, based on the needs and

habits of patients. It is important to tear down the prejudices and fears regarding AI and

understand how it could be beneficial and how we can cope with its perceived or real drawbacks.

10
 ACCEPTED MANUSCRIPT

The biggest apprehension we have is that AI will become so sophisticated that it will surpass

human brain capabilities and eventually will take control over our lives. However, if we succeed

in creating ethical standards, developing measures of success and effectiveness, making it

T
available to the mainstream, and not only to the Ivy League medical institutions, by making AI

P
RI
tools open-source and user-friendly and of proven clinical utility, then societal benefits will

accrue from the use of AI.

SC
Acknowledgements:

NU
The authors would like to thank Professor Ted VanItallie for his excellent comments and
MA
suggestions and for the editorial review of their manuscript. Publication of this article was

supported by the Collège International de Recherche Servier (CIRS) and Canada Research Chairs
ED

to PH. The authors are members of CIRS.

Conflict of interests: None

PT
CE

References
AC

[1] Needham J, Kuhn D, Tsien T-H. Science and Civilisation in China: Vol. 4, Physics and

Physical Technology; Pt. 2, Mechanical Engineering: Cambridge University Press; 1965.

[2] Wiener N. Cybernetics: or Control and Communication in the Animal and the Machine.

Paris: JSTOR; 1948.

[3] World Economic Forum. Global Risks 2015. 2015(10th Edition).

[4] Theofilatos K, Pavlopoulou N, Papasavvas C, Likothanassis S, Dimitrakopoulos C,

Georgopoulos E, et al. Predicting protein complexes from weighted protein-protein interaction

11
 ACCEPTED MANUSCRIPT

graphs with a novel unsupervised methodology: Evolutionary enhanced Markov clustering. Artif

Intell Med. 2015;63(3):181-9.

[5] Rapakoulia T, Theofilatos K, Kleftogiannis D, Likothanasis S, Tsakalidis A, Mavroudi S.

P T
EnsembleGASVR: a novel ensemble method for classifying missense single nucleotide

RI
polymorphisms. Bioinformatics. 2014;30(16):2324-33.

SC
[6] Silverman BG, Hanrahan N, Bharathy G, Gordon K, Johnson D. A systems approach to

NU
healthcare: agent-based modeling, community mental health, and population well-being. Artif

Intell Med. 2015;63(2):61-71. MA

[7] Kalton A, Falconer E, Docherty J, Alevras D, Brann D, Johnson K. Multi-Agent-Based

Simulation of a Complex Ecosystem of Mental Health Care. J Med Syst. 2016;40(2):1-8.

ED

[8] Castaneda C, Nalley K, Mannion C, Bhattacharyya P, Blake P, Pecora A, et al. Clinical

PT

decision support systems for improving diagnostic accuracy and achieving precision medicine. J
CE

Clin Bioinform. 2015;5(1):1.

AC

[9] Hellwege JN, Palmer ND, Raffield LM, Ng MC, Hawkins GA, Long J, et al. Genome-

wide family-based linkage analysis of exome chip variants and cardiometabolic risk. Genet

Epidemiol. 2014;38(4):345-52.

[10] Luxton DD. Recommendations for the ethical use and design of artificial intelligent care

providers. Artif Intell Med. 2014;62(1):1-10.

[11] Bartgis J, Albright G. Online role-play simulations with emotionally responsive avatars

for the early detection of Native youth psychological distress, including depression and suicidal

ideation. Am Indian Alsk Native Ment Health Res 2016;23(2):1-27.

12
 ACCEPTED MANUSCRIPT

[12] Pouke M, Häkkilä J. Elderly healthcare monitoring using an avatar-based 3D virtual

environment. Int J Environ Res Public Health. 2013;10(12):7283-98.

[13] Cornet G. Robot companions and ethics a pragmatic approach of ethical design. J Int

P T
Bioethique. 2013;24(4):49-58, 179-80.

RI
[14] Larson JA, Johnson MH, Bhayani SB. Application of surgical safety standards to robotic

SC
surgery: five principles of ethics for nonmaleficence. J Am Coll Surg. 2014;218(2):290-3.

NU
[15] Knight BA, Potretzke AM, Larson JA, Bhayani SB. Comparing Expert Reported

Outcomes to National Surgical Quality Improvement Program Risk Calculator-Predicted

MA
Outcomes: Do Reporting Standards Differ? J Endourol. 2015;29(9):1091-9.
ED

[16] Simonov M, Delconte G. Humanoid assessing rehabilitative exercises. Methods Inf Med.

2015;54(2):114-21.
PT

[17] Felfoul O, Mohammadi M, Taherkhani S, de Lanauze D, Xu YZ, Loghin D, et al.

CE

Magneto-aerotactic bacteria deliver drug-containing nanoliposomes to tumour hypoxic regions.

AC

Nat Nanotechnol 2016.

[18] Mori M. Bukimi no tani [the un-canny valley]. Energy. 1970;7:33-5.

[19] Destephe M BM, Kishi T, Zecca M, Hashimoto K, Takanishi A. Walking in the uncanny

valley: importance of the attractiveness on the acceptance of a robot as a working partner. Front

Psychol 2015;6:1-11.

[20] Kido T. Genetics and Artificial Intelligence for Personal Genome Service

13
 ACCEPTED MANUSCRIPT

MyFinder: Intimate Community Computing for Scientific Discovery Papers from the AAAI 2011

Spring Symposium 2011:8-11.

P T
Figure legend:

RI
DataTrust Pipeline Flow Integration. Figure from Kirby, E., Tassé, A.M., Knoppers, B.M., Joly,
Y., Ngueng Feze, I., Dalpé, G., Phillips, M.S., Tremblay, J., Hamet, P. (2016). P3G DataTrust –

SC
Developing a DataTrust (P3G-DT) Service for Personalized Medicine Research and Therapeutic
Optimization. Manuscript in preparation.

NU
MA
ED
PT
CE
AC

14