How Artificial Intelligence and Deep Learning are Changing

the Healthcare Industry

1
Kuldeep Raj,
2
Roshni Kumari
Computer Science
and Engineering
Asansol
Engineering
College
Abstract- This research paper will discuss the various In our world today, the direction in which we are headed
technological innovations that have been developed within contains a future that will heavily depend on artificial
the past, are being developed in the present and will be intelligence and deep learning technologies. There are
developed in the future to bring progressive change within currently some technologies that implement artificial
the healthcare industry. In particular, the way artificial intelligence and deep learning that has helped our way of
intelligence and deep learning has brought about new living. The most relatable that we tend to miss as users, is our
methods for making task more efficient, less time cellphones and computers. These devices have programs
consuming, and possibly more accurate. This study will installed that use algorithms to help shift how we receive data.
discuss the different positive and negative effects of these As it relates to our health, we no longer have to calculate all of
technological methods to best see how artificial intelligence our workouts or intake of calories and derive conclusions on
(AI) and deep learning is impacting the health industry in our health based on what our devices have picked up on. For
all areas. With supportive interviews from medical example, there are programs that will tell you based on your
professionals, data scientist, and machine learning diet & body, of what is suggested for you to do to progress.
researchers, the analysis of real-world experiences and On a complex level with physicians using this new approach
scenarios will provide more insight to what is beneficial or of deriving solutions from these technologies come with new
ineffective. The study utilizes extensive literary works to risks, challenges and benefits which will further be discussed
examine in detail the current AI and deep learning new throughout this research.
approach to medicine, the risk and ethical concerns, the
technological challenges from determining the best II. Methodology
algorithms, and myths versus reality surrounded by this This study will use a combination of literature review and
subject. interviews to collect data and gather results directly related to
our thesis. Each stage of our methodology is explained as
I. Introduction follows:
To first establish some preliminary understanding, artificial
intelligence is programming a computer or machine to think and A. Literature Review
draw conclusions similar to humans. Deep learning is a This paper will discuss the various developments of
category under artificial intelligence that uses neural networks technologies centered around health and medication; and risks
– computing algorithms that contain layers of neurons to learn associated with the use of them by relating to scholarly articles
an input of data to provide an output based on its and research reports of researchers and experts in the area of
understanding of it- to learn large amounts of data and solve artificial intelligence and deep learning. Analyzing through
complex problems. these reports will provide a foundation to grow our research on
and will present itself to support the findings from the other
methods used to collect data. References to news articles will
also be used in order to relate our research findings to current
events related to.

B. Interviews
There has been data collected about our topic by interviewing
several individuals who are proficient in the field of artificial
intelligence and health care. The interview process will
involve approximately six experts, 2 medical professionals
and 4 professional researchers.
Figure 1: Deep Learning Network
can bring massive changes in the healthcare industry
This section will cover the cumulative results obtained from
our research methodology outlined in Section II.

A. AI & Deep Learning Technologies Currently Used

The focus of artificial intelligence within deep learning being
used in the healthcare industry can be overlooked at times since
some professionals who are using these technologies know how
to use them but not completely understand what causes the
technology to fall under deep learning. Within the findings of
this research, it was found that there were both simple and
complex usages of deep learning within healthcare that is
occurring at this moment.
 Google developed a combined team at UC San
Francisco, Stanford Medicine, and the University of
 Chicago Medicine to work on an algorithm that
analyzed over 46 billion data points out of 216,000 B. The Future Impact AI and Deep Learning Will
EHRs within two hospitals. Have on Health Care
Due to the head start artificial intelligence and deep learning
Furthermore, if data is not as excessively available, then has in the clinical setting, the future for health care taking
Generative Adversarial Networks are used in order to predict bigger approaches technologically looks to bring promising
rare diseases or unique cases. One Artificial Neural Network results. Scientists and medical professionals are seeking to
is created to generate new data instances from a given data improve the way patients and consumers interact with
set, while another ANN is used to discriminate the data set healthcare systems. This research will take a look at both the
being formed by the first ANN by being given the original set small and grand scale of technologies, ranging from
of data and faulty set of data to determine which is real to wearable technology to cloud computation [19].
provide the first ANN more quality data to generate [5].

 Enlitic had a research team that introduced a tool

which went beyond the combined abilities of
specialized radiologists at identifying lung cancer
nodules in CT images, where the testing
conditions reached a 50% higher detection rate
[5].

Predicting the health condition and providing treatment

before detecting it is much more valuable to both the patient
and physician. In 2006, half of about 4 million
patients hospitalized for preventable hospitalizations
suffered from diabetes and heart problems and cost the
US more than With the transformational outcomes that
have been processed from deep learning algorithms
currently, on a much smaller scale, clinical applications
that are centered around consumer facing technology
can change how patients depend on physicians and
nurses. After researching to determine the larger scale
impact that the future of deep learning will take place
within the healthcare industry, it was discovered that the
main focus can be centered around voice recognition
technologies, cloud computation, robotics surgery,
public health and patient monitoring.
Thankfully the progression that has been made outside of
the healthcare industry with these technologies and areas
of impact has made it more of a reality to see occurring
for where the future is headed, even when it can be more
complex.
 Myths vs Reality
With new technologies on the rise, minimal common
definitions, and the lack of common validity for the best
practices, different myths start rumoring within the field of AI
and the use of big data. This research will identify these myths
in order to shed light to what’s factual, so that the best
approaches can be conducted when applying deep learning to
the healthcare industry. The focused myths that will be
discussed are the need for large volumes of data, the need for
cleaned data, and managing the data from deep learning
technologies requires a lot of money or many data scientists
Figure 2: Change in AI [3]. In reality, none of these are the case for implementing
deep learning within the healthcare industry.
With the transformational outcomes that have been processed
from deep learning algorithms currently, on a much smaller
scale, clinical applications that are centered around consumer
facing technology can change how patients depend on
physicians and nurses. The implementation of chatbots and
virtual assistants in healthcare isn’t used as much as it is in
business offices but based on the global virtual healthcare
assistant market being valued at $376 million in 2018, it’s
looking to increase by over 24% by 2025 [3]. The most
popular use of these technologies includes Alexa, Siri, and
Google Assistant. The benefits from these systems can
Figure 3: Input/output of data processed thru deep
come in the form of providing chronic disease management
learning technologies [3]
programming within patient’s homes, daily availability to
simple triage, and effective ways of accomplishing The last myth that will be discussed relating to big data and
administrative tasks [8]. deep learning is that it requires a lot of money or many data
scientists to manage it. If the situation was to be looked at
After researching to determine the larger scale impact that the financially, then currently the cost of cloud options are
future of deep learning will take place within the healthcare reducing, and healthcare companies can start out looking to
industry, it was discovered that the main focus can be centered invest in open-source technologies. A good team of data
around voice recognition technologies, cloud computation, scientists can become challenging to hire since they are in
robotics surgery, public health and patient monitoring. high demand and require a high salary. So, if the corporation
Thankfully the progression that has been made outside of the is not large, it is suggested to start out using the data scientist
healthcare industry with these technologies and areas of that is hired to the company’s advantage. Use their projected
impact has made it more of a reality to see occurring for skills and assets to focus on what looks probable of
where the future is headed, even when it can be more accomplishing [3].
complex.
Google is currently working with physicians and data
scientists at Stanford University to see how their technology
can bring more change within the clinical setting besides its
current capabilities of transcribing notes. Automatic speech
recognition solutions can be used as a model to develop the
control of handling complex medical diagnosis.
 make within their line of work is not always consistent. The
On the contrary to the myths that were presented the reality type of equipment and systems that are being used can be
and best approaches include, but are not limited to having a different across the healthcare industry. So, the best way to
scope, variety, data flow diagrams, and cyber security [3]. The defend this barrier is by providing training to find irregular
scope for where the data is coming from and how it’s inputs so that modeling outputs are not generating incorrect
analyzed will decrease the chances of being overwhelmed data for healthcare professionals to draw conclusions on. The
with data management. data sets being given have to also be distinctly immense in
size and diverse in quantity from the provider of training the
C. Scientific AI Challenges model and outside of those institutions. If this is conducted,
Artificial intelligence is still progressing and is still fairly new then the challenge can be easier to face when reflecting the
to the healthcare industry in comparison to other technologies differences amongst patients and their health conditions in
that’s been implemented through various businesses or practical situations [12].
organizations within the healthcare industry. However, with
studies still being discovered there are still some adversities to D. Risk and Vulnerabilities
face and limitations that will come across before scientists and There are some clear-cut risks that can be thought of from just
researchers apply more technologies in the real-world for thinking about implementing artificial intelligence and deep
clinical practice. Some of the challenges that are faced include learning into the healthcare industry simply because people’s
differences within the available algorithms, machine learning lives are at stake. This is a more serious situation than others
difficulties, changes within datasets, and generalizations that that may be able to take considerations with a less serious
impact new environments [12]. approach, such as video games, that can leave room for errors
here and there. As it was touched on in the last section, the
The difficulty that physicians will face when finding which risk of the algorithms and technologies being faulty and not
algorithm works excellent for patients is having various results producing the most accurate outputs can lead to misinforming
from executing variable methodologies on a variety of patients with information which will ultimately have a
populations that are made up of multiple characteristics. rippling, devastating effect
Without focusing each AI algorithm on the same type of group
of people, then it will be hard to determine the most accurate Next, one of the biggest risks that comes with implementing
results of which algorithm is best. A good practice to avoid an artificial intelligence and deep learning in the healthcare
approach that won’t provide usable results is to select data and industry is privacy and security. The most immediate
patients being tested from the technologies, from their natural vulnerability that comes with protecting the privacy of a
state [12]. patient, stems from them having to provide the artificial
intelligence systems with so much sensitive data. This data
Machine learning has its own difficulties as well when it can potentially get in the hands of other scientists and
relates to the science of it. Before the application process developers who have to manage the data and technologies
there are situations and considerations that need to be since deep learning technologies and models would be new to
analyzed. clinical practices. Patients have the right to view this as data-
Those key details include changes within datasets, widely sharing which puts them in position to think about loss of
spreading unplanned biases within doctor’s or physician’s privacy or lawsuits to be made
work, estimating how reliable a deep learning model will be,
and looking upon a population with generalizations [12]. In
particular, the changes within datasets that occur overtime can
cause AI and deep learning algorithms to degrade and not be
as efficient. This causes research professionals to be in
position to reexamine the algorithms and determine if they
should only update the model or completely retrain it so it can
do what it was designed to do to the best of its ability.

Throughout this research, there’s a recurring discussion about

generalizations in populations, and this section will lastly
cover the barriers that are faced centered around this. Making
dependable general inferences on a population can be far in
reach because the approaches that doctors and physicians
 E. Ethical Concerns
The risks that were covered in the section above are key
factors to observe in the discussion of what is considered an
ethical approach in order to limit those specified risks. Similar
to the risk being identified, there are some ethical concerns
that can be easily identified and addressed such as informed
consent, the choice for patients to share their data, and
ownership of their data [3]. Some ethical challenges that
require a more in-depth look would be related to the
algorithms such as discrimination and racial bias, corruptible
designs of the machine learning system, and the potential
overpowering authority of a system.

The last ethical challenges that will be discussed are the

ones that require health organizations and businesses to go
beyond the surface. The way health care is provided to
people is already differed by race. When training an
algorithm, careful observation of discrimination has to
occur because past data and results may have been racially
biased. Therefore, if this data is used to train the algorithm
then it is highly likely the algorithm will produce biased
decisions as well [14].

F. Interview Responses
After providing an interview questionnaire to four
professionals, two within the medical field and two
researchers studying artificial intelligence and machine
learning, I’ve been able to receive responses to support the
literary study that was conducted within the research. This
also gives an opportunity to get a better understanding from
those who have valuable professional experience because
their opinions can highlight differences within the research as
well. Since this data is qualitative and not quantitative, this
subsection will discuss the exact responses from the
interviewees. The next section will go further to explain
similarities and differences in perspectives from both sides of
the professionals to suggest important considerations in
regard to topics discussed in this current section. The first
part of this subsection will highlight the responses from the
medical professionals who perform clinical work. Below will
list the questions that were asked specifically to them and
their relative responses:
 Have you ever used AI technology in your area of
work?
o Dr. Ellis: Yes
o Dr. Gray: Maybe
 If selected no or maybe please answer
this: https://novatiosolutions.com/10-common-
applications-artificial-intelligence-healthcare/ : This
link leads to a list of potential uses of AI
technology in the healthcare industry. Please list at
least three
 of which you believe would be most impactful in care and empathy to the patient during
your field and why? encounters, and performing "hands on"
o Dr. Ellis: “I did not select no or maybe.” physical examination of the patient is all
o Dr. Gray: “AI for health monitoring (fit too often being diminished in patient
bits, sleep tracking - this is what patients encounters. They are being replaced by a
have usd to help us make behavioral health reliance on tests, imaging, and patient
changes; more of this would be helpful so surveys (like this) regarding things like
we know what patients are ACTUALLY depression. This reduces the somewhat
doing and how well they are moving subtle but important human aspect of a
toward their goals). AI for review of health practitioner being a healer that gives the
records would be helpful & for system patient a sense of confidence in his/her
analysis (to see what components are most care.”
helpful in patients who make the most
o Dr. Gray: “Negatives could be patients
progress, identify errors, etc)”
using/misinterpreting information, mass
 If selected yes: How has this AI technology availability of individual patient data
impacted the healthcare/medical industry? If it has (patient confidentiality). Benefits are more
not, how can you potentially see this technology information to guide decisions.”
bringing change to the healthcare/medical industry? Next, the responses from the researchers in artificial
o Dr. Ellis: “It has helped in diagnosis and intelligence and deep learning that volunteered to interview
treatment support in my practice. For will be expressed below. Two questions are slightly tailored
example if I diagnose a patient with low to fit their profession because technology is where they are
back pain, my electronic medical record most experienced:
has built in to it the next step an array of  What Artificial Intelligence technology have you
choices such as ordering physical therapy, used or discovered within your field of
back X-rays, medications often used, etc. research/work? (List 1-3 of the ones you find most
valuable)
o Dr. Gray: “AI could help to make big data
o Dr. Johnsen: “That's a pretty broad
more accessible to providers and help
question. Most of my data analytics work
providers make decisions to guide
is done in Python, so the SciPy stack of
individual treatment progress based on
technologies. I also used Unity3D for
more data/info (either from that patient or
interactive graphics.”
from other patients)”
 What are positive and negative effects that you can o Dr. Moayed: “I have used the Chaotic
see occurring if this technology was used in your
Neural Network, Swarm Intelligence, and
field?
Evolutionary Algorithms as well as Deep
o Dr. Ellis: “Positive: greater speed and
Learning. Currently, the most valuable
efficiency in managing patient care in the
technologies in AI are all in the area
case of diagnosis and decision support.
known as Deep Learning and include in (in
Greater objectivity in the grading of visual
no specific order) (1) Transformers (and its
clinical data in the case of biopsies and
predecessors such as Google-developed
pictures as described above. Negative: the
BERT, OpenAI-developed GPT-1,2,3 used
human elements of listening carefully to
for Natural Language Processing), (2)
establish the source of a problem, showing
Generative Adversarial Networks (GANs)
(3) Convolutional Neural Network (CNN)
and (4) Deep Reinforcement Learning
(DRL). These areas are fast-changing and
as we speak new areas of AI are emerging
such as Meta-learning.”
 What are their purposes and functionalities?
o Dr. Johnsen: “Much of the work is data
wrangling, which is elegant using Python.
Unity3D is used more for developing
intelligent agent simulations that you can
interact with.”

o Dr. Moayed: “Deep learning (DL) since

2012, has changed many industries from
speech recognition, machine translation,
context comprehension, to machine vision,
autonomous robotics/driving, and context
generation.”
  How has this AI technology impacted the beings for providing services through face-
healthcare/medical industry? If it has not, how can to-face contact. I haven't seen much AI
you potentially see this technology bringing change impact our daily lives yet, unfortunately.
to the healthcare/medical industry? Much of this is due to various legal issues
o Dr. Johnsen: “I don't think AI has had a surrounding its use, which largely protect
huge impact yet on the medical industry as us from "bad" or "biased" AI being
a whole. I do see fields like Radiology employed. If and when AI systems can
being dramatically transformed in the near perform various roles in the healthcare
future though, because it is so heavily industry reliably, they will be replaced,
based on human analysis of images. and the world will likely be better for it.
Intelligent agents can be and are used for But this is only going to happen when the
teaching purposes. In the much farther healthcare industry can lower costs and not
future, we could see some types of surgery get sued for doing it. The same situation is
being automated through robotics. Perhaps happening now with autonomous vehicles.
we will see some innovation with AI and The industry is almost to the point where
electronic health records, though for any they can let cars drive themselves safely.
given individual, the data is extremely The next steps are lowered costs, and then
sparse. A much greater benefit could be adoption will be immense. The "natural"
gained from AI that helps individuals environment of the human body and
make better choices.” human interaction is a much more difficult
problem for AI, so it's going to take a bit
o Dr. Moayed: “So far, the role of DL and longer for AI systems to become reliable.”
AI has been assistive in helping doctors
and healthcare professional in diagnosis or Dr. Moayed: “Like any other new technology in the last
robotics surgery. One reason is that the hundred years, AI can come with its positive and negative
changes in medical fields need to pass its effects. The most frightening situation I would see is AI being
test of time and be approved through in the hands of a few to create a class of useless people* (term
rigorous procedure by the medical used by Yuval Noah Harari in "Homo Deus: The History of
authority such as FDA. But to my point of Tomorrow"). The positive effects of AI are beyond
view, this will not stop there, and it will imaginations, from regular applications such as automations,
change the medical field dramatically to a robotics to and giving humanity a chance to be independent
point that AI diagnosis will be much more from needing to work to live
accurate than the human diagnosis. There III. Analysis of Results
have been many similar success scenarios This section will analyze the responses from both professional
in the last few years that makes me hopeful groups and draw out qualitative information from each to base
that this is happening sooner than later. the educated probability on their similarities and differences.
Stories such as the impossible Alpha-Go
victory in 2014 in which nobody (even the A. AI and Deep Learning Technologies
most optimistic AI researchers had not Currently Used Analysis
seen that coming) had no expectation that When analyzing both groups of interviewees, there are some
an AI would win a game that literally distinct differences in the technologies being used in their
needs human intuition.” perspective fields. This is likely because of their particular

Figure 4: Robotics Surgery

 What positive and negative effects have you seen
(or can potentially see) with the technology being
used in the situation from the last question?
o Dr. Johnsen: “Greater standardization
amongst the medical community is an
important side-effect of the use of more
AI. This potentially frees up more human
practices and how they have made an impact within each
industry as practitioners and developers. However, one
reoccurring theme that stands out as a similarity between the
doctors and the researchers is the use of electronic health
records to make diagnosis and draw other conclusions. Dr.
Ellis stated that he currently uses it to receive the next steps
for what a patient should do for their current condition such
as ordering physical therapy, back X-rays, and suggest
medications to use.

B. The Future Impact AI and Deep Learning

Will Have on Health Care Analysis
The future impact that artificial intelligence and deep learning
will have on healthcare is limitless. Yes, this sounds touché,
but because we’re still at the early stages of development we
can say that there is so much ahead of us to see and
experience. Based on some important details from the
interviewed professionals that will be discussed in this section
as well, the future of health care is headed in on an innovative
path.

C. Positive and Negative Impact Analysis

Both parties who responded to the question centered around

what are the positive and negative effects that the AI and
deep learning technologies they’ve had experienced with or
can see being used in the future has had on the healthcare
industry.As each answer was read from above in the
interview results, it is seen that there are some promising
benefits that can change the world as we know today for a
better tomorrow. However, there are some distinct ethical
concerns found in the responses that can be related to what
was covered in the literature review that have to be taken
seriously as well. This section will begin discussing both
views of the positive and negative effects to see either
similarities or differences to make an educated analysis.
 IV. Recommendations

From what has been able to be analyzed, I would recommend

that we take into account what every group of people are
saying who play a role in this process of making the
healthcare industry change for the better with the next
technology, which also includes the patients who tend to be
overlooked most times. I see from the literature review when
looking at ethical concerns is that patients don’t get informed
as much as they should and neither do, they have a say until
what’s being developed has already been implemented into
practice.

With that being, said I believe that users of different

applications and patients who will be partaking in various AI
and deep learning led clinical treatments should educate
themselves on the privacy laws and policies, what type of
data they are giving up, and all of the other important
information
they’ll be signing up for before carelessly selecting that they
are okay with what is being done

V. Conclusion
In conclusion, the healthcare industry is still at the beginning
stages in regard to the implementation of AI and deep learning
technologies in practice when comparing where the future is
headed to. There are small usages and complex usages in
today’s time, but as said before it’s not standardized for the
community of physicians and medical professionals. It will
take years, but the amount of time to invest in this is
quality and valuable. This process should not be rushed
because
people’s lives are at stake, and with the proper precautions and
risk evaluations that need to constantly take place then it will
be clear that the benefits outweigh the deficits.

References