How Machine Learning Can
Change the Medical Field
Translation Report
Some Common Uses of Machine Learning
“Most Common Use Cases for Enterprise Machine Learning”, New Gen Apps. [Online]. Available:
https://www.newgenapps.com/wp-content/uploads/2020/04/1_c_fiB-YgbnMl6nntYGBMHQ-1080x675.jpeg. [Accessed: May 7,
2020]
Prepared by: Joseph Haluska
May 22, 2020
Table of Contents
Introduction ................................................................................................................................... 1
Machine Learning Process ........................................................................................................... 1
Machine Learning in Diagnosis ................................................................................................... 3
Imaging .......................................................................................................................................... 4
Precision Medication .................................................................................................................... 4
Epidemic Prediction...................................................................................................................... 5
Conclusion ..................................................................................................................................... 6
Works Cited ................................................................................................................................... 7
Figures Cited ................................................................................................................................. 8
Table of Figures
Figure 1: Uses of Machine Learning in Medicine ...................................................................... 1
Figure 2: Steps of “Explainable” Machine Learning ................................................................ 2
Figure 3: Accuracy of Breast Cancer Diagnosis ........................................................................ 3
Figure 4: Retina Scan Using Machine Learning ........................................................................ 4
Figure 5: Steps in Ethiopian Malaria Prediction ....................................................................... 5
Introduction 1
Introduction
In the field of Computer Science, scientists need to develop algorithms that analyze large sets of
data and perform complex computations. These programs maximize the efficiency of all types of
industries and help them make more sound business decisions. To create these algorithms
computer scientists rely on a method called Machine Learning. Machine Learning allows for the
creation of algorithms that learn and improve as more tests and data are processed [1]. Using
these algorithms businesses can maximize profit, increase efficiency, and cut down on errors.
Every industry is adapting to a new era of technology. This has created a huge new pool of data
and information. With such an abundance of data online, companies are looking for ways to
utilize this information. But there is way too much data for humans to comprehend, examine, and
analyze. Instead, companies have employed Machine Learning to examine patterns within data.
These programs are quicker than humans and can analyze data in much more depth [1]. Machine
Learning is replacing old and outdated methods within industries to create more modern
solutions to society’s issues.
With the current COVID-19 pandemic, the medical industry is demanding modern solutions to
medical issues. Researchers are exploring the use of Machine Learning in the medical field, but
have found no clear application for combating COVID-19 as of now. But applications of
Machine Learning in medicine could play an important role as the pandemic continues or new
pandemics arise. Machine Learning does not even have to directly combat the COVID-19 virus
to benefit the medical field. Using Machine Learning to help address any type of medical issue
frees up resources to deal with COVID-19. Thus, Machine Learning ensures the medical field is
better prepared to deal with pandemics and an increase in hospital admissions. Figure 1
illustrates the current areas within the medical field that are experimenting with Machine
Learning.
Figure 1: The Uses of Machine Learning in the Medical Field
Data adapted from [2]
Machine Learning Process
Machine Learning is a type of algorithm that learns from data, improves over time, and makes
predictions and computations using the patterns it finds within data [2]. These types of
algorithms continue to improve as larger sets of data become available and the processing
Machine Learning Process 2
capabilities of computers increase. As these algorithms improve, their popularity also increases.
At the moment, a very popular method of Machine Learning is Deep Learning.
Deep Learning is very popular because it has matched or exceeded human productivity in many
complex tasks [1]. This algorithm employs a “black box” method of Machine Learning [3]. In
“black box” models an algorithm is first developed by passing a large set of input data into the
algorithm. The algorithm then uses statistical analyses to find patterns within the data [3]. Once
the algorithm analyzes the data, the data is stored within the algorithm for future use. New inputs
are then passed into the algorithm and compared to the data the algorithm has already stored and
analyzed [3]. The algorithm will then output solutions, group similar data, and adjust itself based
on the new inputs. If an output is incorrect, researchers will pass the output back into the
algorithm, allowing the algorithm to update and “learn” from its own mistakes [3].
This type of algorithm is called a “black box” because it hides the mathematical operations
performed to arrive at its conclusions [3]. The program only presents an outcome, with no
explanation of how the outcome developed. Thus, it can be hard to deduce how a program
arrives at a specific output. This lack of a clear “path” to a solution has made it so professionals
are reluctant to adopt and implement Machine Learning. This is especially true in the medical
field because the algorithms are making life or death decisions.
Many medical professionals are afraid that machines cannot accurately make decisions that will
determine whether a patient continues living or not. However, recent clinical testing of Machine
Learning has shown that it is as accurate as humans [4]. Despite this, many are still reluctant to
rely on Machine Learning. This has led to the development of “explainable” Machine Learning
[3]. “Explainable” Machine Learning employs an updated “black box” model. Data is input into
the “black box” and solutions are output in a similar fashion as explained previously. But then
solutions are input into another algorithm that produces an “explainable model” [3]. This model
interprets and explains how the “black box” arrived at its conclusion and translates it to terms
that medical professionals can understand [3]. Then the algorithm displays an interactable
interface that professionals can use to find and correct errors made by the algorithm [3].
“Explainable” Machine Learning means that medical professionals are more willing to begin the
implementation of Machine Learning practices. The use of Machine Learning used in combination
with practicing medical professionals helps the medical field become more efficient and precise. By
using the process of “Explainable” Machine Learning (seen in Figure 2), the medical field will
become better prepared for emergencies like the COVID-19 pandemic.
Figure 2: Steps in “Explainable” Machine Learning
Data adapted from [3]
Machine Learning in Diagnosis 3
Machine Learning in the Diagnosis of Illnesses
Cancer is the most dangerous disease in the world, and breast cancer is the most dangerous form
of cancer for women [4]. Diagnosing breast cancer is typically slow and difficult. But with the
implementation of Machine Learning, breast cancer diagnosis is much quicker and matches the
accuracy of classical diagnosis.
In 2018, researchers set out to develop a breast cancer diagnosis algorithm [4]. To develop this
algorithm, researchers first needed a large set of data. Researchers used a breast cancer data set
given by the University of California Irvine (UCI) to develop their Machine Learning algorithm.
But this data was not perfect. This data set contained outliers and incomplete information that
Machine Learning Algorithms cannot analyze. So, researchers needed to pre-process all the data
before they passed it into a Machine Learning algorithm [4]. Pre-processing is the name of the
operation that occurs when data is put into a “cleaning” algorithm that removes any unusable
information [4]. After researchers pre-processed the UCI data, the “clean” data was put into their
Machine Learning algorithm for analysis.
In a diagnosis algorithm, the algorithm first
identifies features it can use to diagnose an
illness. In the breast cancer diagnosis
algorithm, researchers identified 16 key
features [4]. These are the most important
features that need to be input into the
Machine Learning algorithm when a
patient is going to receive a diagnosis [4].
However, inputting as many features as
possible is important as the algorithm can
learn and find new patterns within new
data that the algorithm analyzes. Once key
features are input into the algorithm, it can
then determine if a tumor is cancerous or
harmless. When researchers tested the Figure 3: Accuracy of Breast Cancer Diagnosis
Data adapted from [4]
algorithm using data from UCI, it was able
to determine if a tumor was cancerous or harmless with 98.5% accuracy [4]. This accuracy is
much higher than the typical accuracy of a radiologist’s diagnosis, as depicted in Figure 3.
This accuracy shows that Machine Learning has a great potential for the diagnosis of different
illnesses. However, this technology is not ready for use in the medical field. The breast cancer
detection algorithm was only tested with data gathered by the University of California Irvine [4].
Researchers must test the algorithm on a more diverse group before the algorithm is safe for use.
As researchers test the algorithm on different groups, it will gather more data. This data will help
the algorithm become more precise and consider different outliers. This testing is very important
because a certified Machine Learning diagnosis algorithm will make diagnosis quicker and less
error-prone. Even though Machine Learning can diagnose illnesses completely by itself, it can
also create images that aid medical professionals in diagnosis.
Imaging 4
Imaging
Imaging is the process of creating visual representations through scanning objects. In medicine,
this practice is often done at a microscopic level and needs extreme precision [2]. Machine
Learning creates images by analyzing light waves and performing computations on these light
waves [5]. Once the algorithm runs, it can then create three-dimensional visualizations that aid
medical professionals in solving problems [2].
In the field of cardiology, the study of illnesses related to the heart, Machine Learning is popular
because of its unique imaging capability. It is mainly used to find issues within blood vessels in
the eye that could stem from heart problems. Machine Learning allows medical professionals to
do this through Optical Coherence Tomography (OCT). OCT is an imaging process that
produces an image of a human retina by using an eye scan. Machine Learning enables this
process by analyzing data from a scan of a human eye. In this eye scan, computers collect data
on wavelengths and light intensity that represent the structure of an eye.
The data produced from the
eye scans is not enough to
produce an image of a retina by
itself. The data must be input
into a Machine Learning
algorithm that finds patterns
within data and performs
computations on the data in
order to construct an image out
of the varying wavelengths and
light intensity that was input
[5]. Once the algorithm
analyzes the data, it can Figure 4: Image of a healthy retina created using a Machine Learning algorithm to
produce an image of a retina, as perform Optical Coherence Tomography (OCT) [1].
seen in Figure 4.
After the Machine Learning algorithm creates an image of a retina, doctors can analyze what is
wrong with a patient’s eye. Depending on the image produced, doctors may deduce whether the
patient’s issues stem from artery disease or diabetes [5]. These different causes enable doctors to
determine what treatment a patient will receive. After the doctors deduce the underlying cause of
a patient’s issue, Machine Learning can help prescribe a personalized treatment that best suits an
individual [5].
Precision Medication
Precision medication is the process of medical professionals prescribing medical treatment
according to an individual’s background and medical history [5]. Race, medication an individual
is currently taking, age, and prior medical history all determine the most effective way to treat a
patient. Currently, Machine Learning techniques provide the groundwork for the future of
precision medicine [5].
Epidemic Prediction 5
A precision medication algorithm is first created by analyzing cases of the same illness among
many different people and the most effective treatment in those cases [5]. Then using a patient’s
background information, the algorithm can compare it to other similar cases and deduce the best
possible treatment.
However, the medical field does not currently have access to enough data to maximize the
efficiency of these algorithms [5]. This is because many people are not comfortable with their
health records being public and potentially exploited by the insurance industry and other
companies. Until the government can relieve these fears, Machine Learnings applications will
remain limited by the amount of data in the public realm. Regardless, precision medication is
vital to creating a more effective medical field. This is especially important as the medical field
looks to prepare for epidemics like COVID-19.
Epidemic Prediction
Epidemics are national and sometimes even global occurrences of the rapid transmission of
infectious diseases [6]. Currently most of the world is facing the COVID-19 epidemic. As of
now, most of the world is in the midst of this epidemic. But there is a possibility that Machine
Learning could help predict future hotspots if we face another crisis similar to COVID-19.
Currently, Machine Learning is analyzing locations in Ethiopia to predict a location’s probability
of a malaria outbreak [6]. The malaria prediction algorithm uses the amount of rainfall, relative
humidity, average temperature, elevation, and the amount of known malaria cases to predict if an
area will face an epidemic [6]. The prediction algorithm uses five years of information on
malaria outbreaks and Ethiopian climate to develop its “black box” algorithm that computes the
likelihood a location in Ethiopia will face an epidemic [6]. The initial “black box” algorithm
needs updated values of each variable every month so that it can maintain the accuracy of its
predictions [6].
The “black box” algorithm is
accompanied by an interface system
that shows why the algorithm
chooses the likelihood it did. This
allows experts to check the algorithm
for faulty results and adjust it if it
makes mistakes [6]. Researchers can
input a location’s elevation,
temperature, humidity, rainfall, and
the number of known malaria cases
into the “black box” to produce an
output. The output experts receive
represents a prediction for the
number of malaria cases the area will
have by the end of the next month
[6]. This process is illustrated in Figure 5. Figure 5: Predictive process of the Machine Learning Algorithm
that predicts malaria epidemics in Ethiopia
Data adapted from [6]
Conclusion 6
This predictive process allows experts to track the likelihood an Ethiopian community will face a
malaria outbreak. Using this information, the government can better prepare these areas to face
an epidemic or can put epidemic prevention measures in place. In turn, this algorithm allows the
government to better prepare for medical emergencies than in the past.
The malaria prediction algorithm cannot apply directly to the COVID-19 crisis. This is because
Malaria is an infection, not a virus, and COVID-19 has already reached the status of an
epidemic. However, this principle can be applied to try and predict hotspots for future outbreaks
and epidemics. The “black box” must change to interpret different variables like population
density rather than elevation, but it remains a possibility that Machine Learning can predict
hotspots and a wide variety of epidemics. This is just one of the many ways that Machine
Learning is preparing the medical field for the future.
Conclusion
Machine Learning is a popular method of algorithm implementation that can improve the
efficiency of many everyday processes within different fields. In the future Machine Learning
may be used along with practicing professionals to create more efficient and precise work
environments. In the medical field, it will help contribute to saving lives and easing the burden
on the health care system. As of now, Machine Learning has not seen enough testing for many
medical professionals to feel safe utilizing it in the medical field. On top of this, the public’s
anxiety regarding the spread of private information is holding back the development of Machine
Learning. Until Machine Learning can alleviate the public’s fears and prove itself to medical
professionals, its real-life medical applications will remain limited.
Despite the skepticism some may have for Machine Learning, its applications remain too
important to disregard. In the medical field, not only does it accelerate some already in place
practices, but it has helped develop new ways to solve medical problems. The ability to diagnose
patients more quickly, prescribe individualized treatment, create detailed medical images, and
predict epidemics makes Machine Learning a vital component for the future of the medical field.
The further use and testing of Machine Learning gives the medical field an opportunity to
elongate patients’ lives and increase patients’ quality of life. With Machine Learning in the
medical field there is a better chance to create a future filled with more positive outcomes.
Works Cited 7
Works Cited
[1] S. Feng and J. Boyd-Graber. “What can AI do for Me?: Evaluating Machine Learning
Interpretations in Cooperative Play,” in the 24th International Conference on Intelligent
User Interfaces (IUI ’19), March 17–20, 2019, Marina del Rey, CA, USA [Online]. New
York, NY, USA: ACM, 2019. Available: Engineering Village,
https://dl.acm.org/doi/proceedings/10.1145/3301275 . [Accessed: April 28, 2020].
[2] T. S Toh, et al. “Looking Beyond the Hype: Applied AI and Machine Learning in
Translational Medicine,” in EBioMedicine, vol. 47, p. 607-615. September 1, 2019.
[Online]. Available: https://doi.org/10.1016/j.ebiom.2019.08.027. [Accessed May 5,
2020]
[3] A. Holzinger, et al. “Causability and Explainability of Artificial Intelligence in
Medicine,” Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery,
vol. 9, no. 4. April 2019. [Online]. Available: Engineering Village,
https://doi-org.ezproxy.lib.calpoly.edu/10.1002/widm.1312. [Accessed: May 8, 2020].
[4] N. Khuriwal and N. Mishra. “Breast Cancer Diagnosis Using Adaptive Voting Ensemble
Machine Learning Algorithm” in IEEMA Engineer Infinite Conference, March 13-14,
2018, New Delhi India [Online]. IEEE, 2018, Available: IEEE, https://doi-
org.ezproxy.lib.calpoly.edu/10.1109/ETECHNXT.2018.8385355. [Accessed: April 30,
2020].
[5] S. Niazi, et al. “Cardiovascular Care in the Era of Machine Learning Enabled
Personalized Medicine,” in the 34th International Conference on Information
Networking, ICOIN 2020, January 7-10, 2020, Barcelona, Spain [Online]. Danvers, MA,
USA: IEEE, 2020 Available: IEEE, https://doi.org/10.1109/ICOIN48656.2020.9016426.
[Accessed: April 28, 2020].
[6] B. E. Chekol and H. Hagras, "Employing Machine Learning Techniques for the Malaria
Epidemic Prediction in Ethiopia," 10th Computer Science and Electronic Engineering
(CEEC), September 19-21, 2019 Colchester, United Kingdom [Online]. IEEE, 2019.
Available: IEEE, https://doi-org.ezproxy.lib.calpoly.edu/10.1109/CEEC.2018.8674210.
[Accessed: April 30, 2020].
Figures Cited 8
Figures Cited
[1] “Optical Coherence Tomography (OCT),” Apple Cross Eye Clinic, 2016. [Online].
Available: http://applecrosseyeclinic.com.au/procedures/optical-coherence-tomography-
oct/. [Accessed May 9, 2020].
