Volume 7, Issue 9, September – 2022 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Research of Artificial Intelligence in Medicine

Debashish Padhi (Author)
B.Tech, Department of Information Technology
Odisha University of Technology and Research
Bhubaneswar, India

Abstract:- AI is transforming the world in every walk of technologies. Majority of these technologies affect healthcare
life. It is collection of technologies that in future will directly but they differ by certain ways for particular tasks.
probably rule the entire world. AI is a technology that The technologies that are manipulated to behave like human
thinks like human, learns like human but performs like brain in terms of thoughts, engagement, adaptation, sensible
super-human. In this paper, we will discuss about the feelings, etc. are referred as AI. Alan Turing first brought this
application of AI in healthcare industry and all the pros question to spotlight that, can AI think like humans? [20].
and cons it is having, starting from decision making to Logic theory machine was the 1st AI program that can copy
integrating information, analysing data and use of its own the style of humans in problem solving aspects [21]. Due to
results to improve itself. The technologies like Magnetic this kind of thinking abilities, game like chess can be played
Resonance Imaging (MRI), Computed Tomography Scan by AI [24, 25]. AI plays a vital role in human life not only in
(CT scan), are a part of AI that’s been widely used in healthcare industry but also in social media, technical
healthcare industry. The use of automated classification gadgets, etc. AI is a hugely growing field now days for its
techniques like Machine Learning and Natural Language opportunities in coming future [27].
Processing has proved greater accuracies than manual
classification. AI’s aid the physicians in countless ways in III. AI IN HEALTHCARE INDUSTRY
health care industries. There are certain limitations and
demerits of AI too. AI can’t be trusted since it’s a According to history, AI hands on the medical field
technology that can be tampered. Privacy of patients is mostly around 1960s and 1970s (the term was coined by Jhon
also a matter of concern. There is lack of transparency in Mc Carthy in 1995) [19]. Healthcare industry is aided by the
performing certain jobs. Mainly certain diseases are technologies of AI. AI and the field of robotics are
focused only now days. The ways to improve are discussed enormously developing so as to benefit the healthcare
here and new ideas are proposed for its betterment and industry. In field of medicine, AI got a lion’s share of work,
fast growth. in aiding health practitioners to ease tasks, figure out complex
procedures, etc. [31]. AI eases the tasks of physicians in
Keywords:- Artificial Intelligence; health; machine taking healthcare actions by providing them information from
learning; deep learning; artificial neural network; MRI; CT journals, textbooks, previous records [8, 9]. Modern
scan; supervised learning; semi-supervised learning; technologies enhance the AI abilities to self-correct it and
medicine. learn from mistakes to become more precise in decision
making [10].
I. INTRODUCTION
IV. APPLICATIONS OF AI
This paper provides a wholesome study of AI since it
has been in the healthcare industry. Various technologies of AI is performing in different aspects of healthcare or
AI have been implemented in the medical field such as medicines like clinical, surgical, therapies, disorders and
magnetic resonance imaging (MRI), Computed tomography predictive practices. AI is also doing crucial works in
scan (CT scan), which are one of the most used product of AI. decision making and patient diagnosis. Here AI is making an
AI has been a great help for physicians since long. Machine enormous impact. The very best thing of AI is that it can
learning algorithms have been studied here. Decision making digest large volumes of dates across different platforms and
tendency of AI have been complimented by many, for good. it can analyse them to determine the diseases and provide
Still AI has a long way to go for its proper full implementation decisions precisely [28, 30]. Ai is essentially needed in early
in healthcare industry. Rules and regulations regarding its detection and diagnosis works. AI and robotics are getting
transparency and trust issues are providing constraints. Ways developed for this application especially. AI can now do
to improve AI in near future have been discussed later. perform human works in a much more effective way with low
cost. Diseases like cancer are hard to detect without the help
II. WHAT IS AI of AI. Ai especially detects these kind of harmful diseases
quickly reliably in a more accurate way. Ai makes the works
AI stands for artificial Intelligence. As the name of the routine check-up automatic; otherwise there have to be
suggests it’s an artificial human brain. In real world, when it people to do medical practice of patients.
comes to AI, it is seamlessly the integration of pack of

IJISRT22SEP167 www.ijisrt.com 60
Volume 7, Issue 9, September – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
V. MACHINE LEARNING (ML) C. Semi-supervised Learning
There is a midway between supervised and
ML is a way of dealing with technologies by putting unsupervised learning, named as semi supervised learning. It
data through the help of models in a statistical way. It’s the is basically follows the trend of disease pattern and makes the
most use form of AI [1]. ML with the help of data through bridge where data are missing. It combines small amount of
different platforms, help to predict which kind of medicine or labelled data with huge amount of unlabelled data. It is a self-
treatment will be best suitable for a particular patient based in training model of Machine Learning.
his health care status. It basically provides the ways based on
previous treatment data. For it to work, it majorly requires From the above categories of ML, most trustworthy
enormous amount of patient dataset to provide the treatment among all is supervised learning. Neither it makes the data
precisely [2]. like semi-supervised learning nor does it try to find
informative data like unsupervised learning.
VI. TYPES OF AI
VIII. ARTIFICIAL NEURAL NETWORK (ANN)
AI is majorly divided into two categories based on their
approaches to work [11]. It is a type of computing system inspired from
biological neural networks. It is also known as stimulated
A. Machine Learning Technique neural networks or the neural nets. Their name and structure
It is the AI which make of ML techniques to make are inspired by the human brain, mimicking the way that
decisions in a more statistical way by dealing with vast biological neurons signal to one another [39]. It basically
amount of structured data. In healthcare industry the ML works like our brain’s neural network system. ANN mimics
techniques gives the most probable output based on the input the process of working of our brain’s neural system [22, 23].
dataset provided of patients character. ML techniques use It provides a probability of an event or disease to happen in
structured data [3]. an individual. ANNs are created by programming regular
computers to behave as though they are interconnected brain
B. Natural Language Processing Technique cells. ANN would be provided by the training set that would
As the name suggests, this model of AI tries to provide thousands of images tagged as a particular thing to be
understand, analyse, manipulate and generate human determined in future, so the network would begin to learn
language [17]. It is a type of the AI which extract information [14]. Once it has been trained with the significant amount of
from unstructured data (for example clinical prescriptions, data, it will try to classify future data based on what it thinks
journals, etc.) [26]. it’s seeing (or hearing, depending on the data set) throughout
the different layers of neurons.
VII. CATEGORIES OF ML
IX. DEEP LEARNING
A. Supervised Learning
It is a Machine Learning method. It is the basic way how As the name suggests, it simplifies how far the AI is
ML works. Takes input of labelled dataset (i.e. how to deal processing the data for providing the most accurate result.
with patient’s disease) and then training process is done. After The name given to the length of processing is “deep”. So it is
it get test data to predict the output. The aim of a supervised known as “deep learning”. The term “deep” signifies the
learning algorithm is to find a mapping function to map the number of neural networks connected with each other so as
input variables with the output variables. It is a mathematical to go through each of the probability of upcoming outcome
model which has the inputs and the desired outputs. and proving the desired outcome. Deep learning is actually a
part of Artificial Neural Network [4, 5]. ANN has actually
Types of supervised-learning algorithms include one neural network connected to several neuron networks
1. Classification (say 10 for example) and if those neuron networks (as
2. Regression example 10) are again connected to several networks and this
chain repeats for several times, then this chain is between the
B. Unsupervised Learning input variable and output data and known as deep learning
It is a little advanced technologically. It scoops the [16]. It is regarded as the most complex form of ML. It helps
desired information from hand written medical notes. It in detecting diseases like cancer etc. from scanned images in
analyses and clusters unlabelled data sets. These algorithms radiology by taking the very insignificant details /data which
discover hidden patterns or data groupings without the need is beyond the human eye to detect. Classical neural network
for human intervention. Unsupervised machine learning, uses is different from deep learning which in turn is a highly
machine learning algorithms to analyse and cluster unlabelled developed version of classical neural network. Deep learning
datasets. Unsupervised learning method takes the data and has more layers unlike classical neural network [13].
searches the hidden pattern in them and groups the data into
sub-groups according to the pattern. When new data comes, A. Types of Deep Learning:
unsupervised learning method detects for which pattern is  Learning rate decay
being used there and puts the newest data into that sub-group  Transfer learning
[12].  Training from scratch
 Dropout

IJISRT22SEP167 www.ijisrt.com 61
Volume 7, Issue 9, September – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
X. TEXT MINING XIII. LIMITATIONS OF AI IN MEDICINE

It is a technology of AI that works on Neural Language A. Need human surveillance

Processing (NLP). It merely works as extracting hidden texts It doesn’t matter how smart AI becomes, there’s still lot
from different texts sources (given texts). It is similar to data of dilemma for its functioning.
mining, which extracts data from pile of data. Text mining Still there are lot of rules and regulations for AI so it can’t be
can play a vital role in healthcare industry for processing led to operate on its own.
medical information. So the human hand is still in need [38].

XI. DATA MINING B. May overlook social variables

Social factors may be neglected by AI while making
This is an AI algorithm that discovers the data from the decisions. For example, for a specific diagnosis, AI can show
pile of data unlike machine learning that predicts the outcome proper care centres but may not consider the person’s
based on the data set. While Machine Learning outcomes are economic or social restrictions and personalised preferences
the predictions of the possibilities of the events (disease) to [38].
happen on the other hand, data mining focuses on discovery
of unknown things from the given data. Many Machine C. Unemployment
Learning methods are used in Data mining technique. Data History has proven, whenever technology evolves,
mining in turn also uses many machine learning techniques. unemployment evolves simultaneously.
Professionals who got trained for years may be displaced
XII. PROBLEMS IN AI cause of AI. Their time and money may get wasted [38].
So factors like social ramifications have to be considered
AI has given subtle examples of its supremacy but still before implementation of AI.
it has a long way to go before termed as fully reliable. There
are vast amount of problems among which few are discussed. D. Inaccuracies are still possible
As AI survives on data, so data set should be large
A. Disease Focus enough for proper diagnosis [38]. If less data (limited data)
Though AI is getting better day by day, but it only available then outcomes of AI can be treated as miss-
focuses on certain problems (medical conditions) in health diagnosis. Limited data can be caused by many factors.
sector only, like cancer, neurovascular diseases and Inaccuracies are highly possible while prescribing a particular
cardiovascular diseases [15]. new type of medicine as patient diagnosed data will be very
less regarding that medicine.
B. Trust
After all AI is certain technology which can be E. Accountability
tampered maliciously. Doctors may not be able to build the There are certain requirements for companies using AI
trust with the data provided by AI cause of possible systems. The medical organisation will be accountable if any
susceptibility [33]. harm is caused by the AI system they handle or work with
[34].
C. Biasing If faults will be done, then accounting a computer system is
Can provide results or AI outcomes can be based on in vain.
racial views. Race can be considered while deciding
outcomes for criminal behaviour [33, 37]. XIV. IMPORTANT TO KNOW BEFORE WORKING
WITH AI
D. Data Biasing
AI is trained by enormous data set. Problem lies if the A. Transparency
data set used doesn’t reflect the targeted population. Data may With proper data set, deep learning model of AI can
be in sufficient for AI to provide wholesome result, like if produce promising outcomes, but deep learning model is full
there is a minority with less health support then their data will of complexities and explaining some medical professional
not be considered [35]. about how the outcome came is very difficult or nearly
impossible, which cuts off AI’s transparency [33].
E. Regulations of AI
Implementing AI technologies in imaginary world and B. Personal
implementing them in real world have tons of difference. In It is very vital to protect the privacy of an individual. It
real world due to the rules and regulations formed on AI is patient’s autonomy and personal identity. Protecting
technologies, its way difficult to improve its safety standards person’s personal information is not that promising about AI
[32]. There should be new laws or old laws should be as, at last it is a technology which can be tampered in a wrong
amended to make the implementation of AI in real world [41]. way [36].

IJISRT22SEP167 www.ijisrt.com 62
Volume 7, Issue 9, September – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
C. Benefits of AI XVII. PERSONAL VIEWS
AI is relentlessly potential for betterment of health care
system. Clinicians can use AI for getting rid of routine check- AI is undergoing pretty much in a decent rate of
ups and those can be automated by AI and clinicians can get development. The infrastructure of AI has to be developed up
time for more important things in their schedule [7]. to the mark so that it can be implemented in other areas too.
The algorithms of AI are not that strong till now to produce
D. Patient Engagement the exact desired result in an operation. Developers have to
Surveys have reportedly said that “patient engagement” widen the area of focus of AI into many other diseases also.
is a very crucial part in patient’s own wellbeing. More interest For replacing the human doctors, AI has to indulge not only
shown by patients for their own health, better their health in assisting the physicians but also have to perform the jobs
becomes [6]. of physicians without involvement of any human hand. AI’s
interaction with patients is a must for talking AI to a next
E. Automation of non-crucial works only level.
Automation of direct patient contact is very unlikely,
which is good news for medical practitioners. Only works, XVIII. CONCLUSION
related to pathology, radiology, etc. that deals with
information will be automated. AI has been able to impress people for many years. In
healthcare industries it is assisting healthcare physicians in
F. Clinical decision making many activities. AI technologies like Machine Learning,
As AI is developed by feeding data, so it can make Natural Language Processing are performing more than
better clinical judgements by analysing tons of data. decent amount of jobs to aid the physicians. Machine
According to Bennett and Hauser, algorithms can be Learning methods like supervised learning, unsupervised
beneficial for making clinical decisions by accelerating the learning and semi-supervised learning are taking technology
process [29]. This will positively impact the cost of healthcare of AI into a new level. Deep learning methods though making
services also. There clinical technicians can be aided by AI AI complicated but is performing up to the mark in most
technologies and AI can make cost of healthcare services cases. With the help of AI technologies like data mining we
cheaper. are now able to discover unknown data from the big data.

XV. AI IN SPECIFIC AREAS After studying all of its pros and cons, the near future of
AI is not that optimistic because of many reasons. Privacy
A. Early detection and diagnosis of stroke issues concerns the most among all. After all it is a
AI are implemented in smart watches, phones etc. to technology, can be tampered. In today’s world there are lots
track human body movement. Before the period of heart of rules and regulations on AI in not only healthcare sector
stroke, movement of patient will be significantly changed. At but in many fields, so implementing the technologies brings
that point of time AI alerts for a stroke. AI algorithms are the biggest hurdles. As of AI no transparency is there. What
developed to detect a stroke before it happens to save goes inside the deep learning layers is hardly explainable for
patient’s life [18]. every output. AI is still now good at only certain diseases, like
cancer, cardiovascular and neurovascular diseases. It has to
B. Patient data and diagnosis be developed for many other diseases and for unknown future
AI is feed by the data set so, through the help of medical diseases. AI does a good job in detecting some particular
big data, AI algorithms can show clinicians the similarities diseases in very preliminary stage but not for all. As of now,
generated from diagnosis, screening, CT scan, MRI etc. AI doctors don’t seem to replace the human doctors quite for
among particular disease pattern to make clinicians aware for a long time.
that.
REFERENCES
XVI. WAYS FOR IMPROVEMENT OF AI
[1]. Deloitte Insights . State of AI in the enterprise . Deloitte
 Algorithms can be developed to make AI question human , 2018 .
being when there is lack of data in a report instead of www2.deloitte.com/content/dam/insights/us/articles/47
bridging the gap by approximate output to finalise the 80_Stateof-AI-in-the-
result. enterprise/AICognitiveSurvey2018_Infographic.pdf .
 To improve the transparency, algorithms may be [2]. Lee SI , Celik S , Logsdon BA et al . A machine learning
developed so that AI robots will answer physicians’ approach to integrate big data for precision medicine in
questions regarding the generated output. acute myeloid leukemia . Nat Commun 2018 ; 9 : 42 .
 The biasing of AI on the basis of race can be eliminated [3]. Sordo M. Introduction to neural networks in healthcare
by feeding AI with the past criminal records from the . OpenClinical , 2002 .
crime branch in detecting the person’s criminal record. [4]. Fakoor R , Ladhak F , Nazi A , Huber M . Using deep
 AI should practice by carrying out real type medical learning to enhance cancer diagnosis and classification
operations on a dummy human body so that at certain . A conference presentation . The 30th International
needs AI doctors will be able to replace human doctors. Conference on Machine Learning , 2013 .

IJISRT22SEP167 www.ijisrt.com 63
Volume 7, Issue 9, September – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[5]. Vial A , Stirling D , Field M et al . The role of deep [21]. Gladwin LA. Alan Turing, Enigma, and the breaking of
learning and radiomic feature extraction in cancer- German machine ciphers in World War II. Available at:
specific predictive modelling: a review . Transl Cancer Alan Turing, Enigma, and the breaking of German
Res 2018 ; 7 : 803 – 16 . machine ciphers in World War II. Available at:
[6]. Davenport TH , Hongsermeier T , Mc Cord KA . Using https://www.
AI to improve electronic health records . Harvard archives.gov/files/publications/prologue/1997/fall/turin
Business Review 2018 . https://hbr.org/2018/12/using- g.pdf. Accessed 15 December 2018.
ai-to-improve- electronichealth-records . [22]. Newell A, Simon HA. Current developments in
[7]. McKinsey Global Institute . A future that works: complex information processing. Technical report P-
automation, employment, and productivity . McKinsey 850. Santa Monica, CA: Rand Corporation.
Global Institute , 2017 . [23]. Newell A, Shaw SC, Simon HA. Elements of a theory
www.mckinsey.com/∼/media/mckinsey/featured%20i of human problem solving. Psychol Rev
nsights/ 1958;65(3):151–66.
Digital%20Disruption/Harnessing%20automation%20 [24]. Shannon CE. Programming a computer for playing
for%20 a%20future%20that%20works/MGI-A-future- chess. Philos Mag 1950;41(314):256–75.
that-worksExecutive-summary.ashx [25]. Samuel AL. Some studies in machine learning using the
[8]. Murdoch TB, Detsky AS. The inevitable application of game of checkers. IBM J 1959;3(3):210–29.
big data to health care. JAMA 2013;309:1351–2 [26]. Weizenbaum J. ELIZA—a computer program for the
[9]. Dilsizian SE, Siegel EL. Artificial intelligence in study of natural language communication between man
medicine and cardiac imaging: harnessing big data and and machine. Commun ACM 1966;9(1):36–45
advanced computing to provide personalized medical [27]. Hamid S. The opportunities and risks of artifcial
diagnosis and treatment. Curr Cardiol Rep 2014;16:441 intelligence in medicine and healthcare [Internet]. 2016
[10]. Pearson T. How to replicate Watson hardware and [cited 2020 May 29]. http://www.cuspe. org/wp-
systems design for your own use in your basement. content/uploads/2016/09/Hamid_2016.pdf
2011 https://www.ibm.com/ [28]. Liao H, Tang M, Luo L, Li C, Chiclana F, Zeng X-J. A
developerworks/community/blogs/InsideSystemStorag bibliometric analysis and visualization of medical big
e/entry/ibm_ data research. Sustainability. 2018;10(1):166
watson_how_to_build_your_own_watson_jr_in_your_ [29]. Bennett CC, Hauser K. Artifcial intelligence framework
basement7? lang=en (accessed 1 Jun 2017) for simulating clinical decision-making: a Markov
[11]. Darcy AM, Louie AK, Roberts LW. Machine Learning decision process approach. Artif Intell Med.
and the Profession of Medicine. JAMA 2016;315:551– 2013;57(1):9–19.
2 [30]. Cho B-J, Choi YJ, Lee M-J, Kim JH, Son G-H, Park S-
[12]. James G, Witten D, Hastie T, et al. An introduction to H, et al. Classifcation of cervical neoplasms on
Statistical Learning with applications in R. First colposcopic photography using deep learning. Sci Rep.
Edition: Springer, 2013. 2020;10(1):13652
[13]. Goodfellow I, Bengio Y, Courville A. Deep Learning. [31]. Coeckelbergh M. Health care, capabilities, and AI
First Edition: The MIT Press, 2016. assistive technologies. Ethical Theory Moral Pract.
[14]. Kantor P. Foundations of statistical natural language 2010;13:181–90
processing: MIT Press, 1999:91–2. [32]. Indonesia R. Undang-undang Republik Indonesia
[15]. Marr B. First FDA approval for clinical Cloud-Based nomor 36 tahun 2009 tentang Kesehatan. Jakarta
Deep Learning in Healthcare. 2017. Republik Indones; 2009.
https://www.forbes.com/sites/bernardmarr/2017/ [33]. Guan J. Artificial intelligence in healthcare and
01/20/first-fda-approval-for-clinical-cloud-based-deep- medicine: promises, ethical challenges and governance.
learning-inhealthcare/#7a0ed8dc161c (accessed 1 Jun Chin Med Sci J. 2019;34:76–83,
2017). http://dx.doi.org/10.24920/003611.
[16]. Ravi D, Wong C, Deligianni F, et al. Deep Learning for [34]. Le Douarin Y, Traversino Y, Graciet A, et al.
Health Informatics. IEEE J Biomed Health Inform Telemonitoring and experimentation in telemedicine
2017;21:4–21. for the improvement of healthcare pathways (ETAPES
[17]. Castro VM, Dligach D, Finan S, et al. Large-scale program). Sustainability beyond 2021: what type of
identification of patients with cerebral aneurysms using organisational model and funding should be used?
natural language processing. Neurology 2017;88:164–8 Therapies. 2020;75:43–56,
[18]. Villar JR, González S, Sedano J, et al. Improving human http://dx.doi.org/10.1016/j.therap.2019.12.009
activity recognition and its application in early stroke [35]. Akmal A, Greatbanks R, Foote J. Lean thinking in
diagnosis. Int J Neural Syst 2015;25:1450036. healthcare – findings from a systematic literature
[19]. Childs M. John McCarthy: Computer scientist known network and bibliometric analysis. Health Policy (New
as the father of AI. York). 2020;124:615–27,
https://www.independent.co.uk/news/obituaries/john- http://dx.doi.org/10.1016/j.healthpol.2020.04.008.
mccarthy-computerscientist-known-as-the-father-of-ai- [36]. Reddy S, Allan S, Coghlan S, et al. A governance model
6255307.html. Accessed 15 June 2018. for the application of AI in health care. J Am Med
[20]. Turing A. Computing Machinery and intelligence. Inform Assoc. 2020;27:491–7,
Mind 1950;49:433–60. http://dx.doi.org/10.1093/jamia/ocz192

IJISRT22SEP167 www.ijisrt.com 64
Volume 7, Issue 9, September – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[37]. Gijsberts CM, Groenewegen KA, Hoefer IE, et al.
Race/ethnic differences in the associations of the
Framingham risk factors with carotid IMT and
cardiovascular events. PLoS One. 2015;10(7):
e0132321. doi:10.1371/journal.pone.0132321
[38]. https://drexel.edu/cci/stories/artificial-intelligence-in-
medicine-pros-and-cons/
[39]. https://www.forbes.com/sites/bernardmarr/2018/09/24/
what-are-artificial-neural-networks-a-simple-
explanation-for-absolutely-anyone/?sh=3f1b840c1245
[40]. Caliskan, Aylin; Bryson, Joanna J.; Narayanan, Arvind
(2017-04-14). "Semantics derived automatically from
language corpora contain human-like biases". Science.
356 (6334): 183–186. arXiv:1608.07187.
Bibcode:2017Sci...356..183C.
doi:10.1126/science.aal4230. ISSN 0036-8075. PMID
28408601. S2CID 23163324.
[41]. Wang, Xinan; Dasgupta, Sanjoy (2016), Lee, D. D.;
Sugiyama, M.; Luxburg, U. V.; Guyon, I. (eds.), "An
algorithm for L1 nearest neighbor search via monotonic
embedding" (PDF), Advances in Neural Information
Processing Systems 29, Curran Associates, Inc., pp.
983–991, retrieved 2018-08-20

IJISRT22SEP167 www.ijisrt.com 65