Professional Documents
Culture Documents
AI AR Based Skin Rash Diagnosis System
AI AR Based Skin Rash Diagnosis System
net/publication/320703991
Artificial Intelligence and Augmented Reality based Skin Rash Diagnosis System
CITATIONS READS
0 2,120
4 authors:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Sneha kasetty sudarshan on 29 October 2017.
B. Chicken pox
Chickenpox is not just a skin disease but, is
very serious and can lead to damage of
important body organs like liver, stomach. It
is highly recommended for everyone to take
vaccination to avoid chickenpox. It is very
Fig. 1: Order of concepts Reality and Virtuality
contagious and very likely to occur at least
Augmented reality differs from its once in a lifetime to those who haven’t been
commonly known “relative”, virtual reality vaccinated. It can be identified by symptoms
(VR). VR creates a 3D world completely like itchy skin rashes with blister-like
detaching the user from reality. There are appearance.
two respects by which AR is unique: users How it spreads:
● From mother to baby during
do not lose touch of reality and it puts
information into eyesight as fast as possible. pregnancy or nursing.
● Through airborne disease carriers.
These distinctive features enable AR to
● Through skin contact.
become a driving force in the future of
● Through infected person’s saliva.
medicine.
● Through contact with contaminated
The Remainder of the paper is organized as objects.
follows. Section II discusses the different Dangers of chicken pox:
rash diagnosis. In chapter III we have In most cases, people infected with
declared the existing systems which are chickenpox recover completely, but for
available for medical diagnosis. Chapter IV people with weak immune systems, if could
we give an overview of the system. Chapter be a life threat. It is difficult to differentiate
V we have described our proposed System between chicken pox and acne as both have
Algorithm. chapter VI we have enlisted the similar effects on skin.
How to differentiate between chicken pox company aims at increasing their sales for
and acne? the cosmetic products by suggesting
In the early stages, the skin rash caused by cosmetics based on their facial problems. It
chickenpox often appears like the red rash detects the skin damage based on the fine
formed due to acne but, the spots will lines and the changes in the texture. The
develop into watery blisters in case of main drawback of this application is that it
chickenpox. These blisters spread very cannot detect any dangerous moles or other
quickly and appear all over the body unlike skin diseases.
acne. After four or five days, these blisters
get old and turn into scabs with a crusty Another interesting application named
outer layer. Mostly, these blisters first occur DoctorMole was recently developed, which
on chest and stomach. Fever is the main can differentiate between a normal mole and
symptom that differentiates chickenpox a mole caused due to skin cancer. In this
from acne. application, the users can take a picture of
their suspicious mole and input it to the
C. Sunburn application which decided whether it is a
When a person is exposed to severe sunlight normal mole or a skin cancer mole [6]. The
or UV rays, it results in a sunburn which formula used behind this is that, if it is a
causes serious skin damage and makes the normal mole, then it will have symmetric
skin too hot to touch. It can be identified by dimensions otherwise, it can be a cancer
red, itchy, painful skin. Sometimes, it may mole. So, depending on the dimensions of
lead to blisters. the mole, the application alerts the user
Dangers of sunburn: whether it is a sign of skin cancer. The main
It can not only be painful, but also result in drawback of this application is that the
dangerous skin diseases like skin aging, pictures and results cannot be exported from
DNA damage and increases the risk of skin the application to get doctor’s advice or
cancer. If a person suffers five or more consent. These results cannot be substituted
sunburns, then there is a high chance of for a physician’s final verdict and should be
increasing the risk of lifetime melanoma by diagnosed properly.
about 80%.
Similarly, there is another application called
III. EXISTING SYSTEMS DermaCompare which can detect signs of
skin cancer just with the help of a picture of
Implementation of augmented reality for the mole [7]. Unlike the previous one, this
skin diagnosis can revolutionize the application uses augmented reality with big
diagnosis procedure. There are very few data technology to compare the existing
applications which have used augmented mole with previous cases of suspicious
reality to study a person’s skin through the cancer moles. This application beats the
camera. In this section, we have discussed previous DoctorMole because it is easy to
about few such already existing applications. use and the pictures can be shared outside
First is Olay skin advisor which, tells the the application. It is very user-friendly and
age of a person’s skin by analyzing the can be used on smartphones and google
picture of their face. It not only tells the age, glass.
but also focuses on specific areas of the face All the above-mentioned applications
and tells up to what extent they have been concentrate only on a single use-case, but
damaged. Through this application, the cannot differentiate between different skin
problems. Our application focuses on For example, detecting some
providing advice to the users based on the cancerous cell in an image [2].
observed skin problem and suggest them on A Medical decision made by an AI machine
how to proceed for can diagnosis based on are not perturbed by causes that are specific
the severity. to humans like stress, reduced attention and
fatigue. It is efficient, can be repeated,
IV. PROPOSED SYSTEM OVERVIEW superior speed and can store a large amount
of data. These systems can select and
The 2 main technologies which we generate important data which will be very
researched for the rash diagnosis are useful for physicians [2].
Artificial Intelligence and Mobile
Augmented Reality. In this paper, we use Artificial Intelligence
(AI) technology support for decision making
A. Artificial Intelligence of the next steps required in diagnosing a
Artificial intelligence is the ability for rash.
machines to possess human like intelligence
and mimic the human brain [8]. Artificial Overview of AI Mechanism for Medical
intelligence has advanced over the past years Predictions
and helps to solve many challenging In this paper, we discuss two AI approaches
problems in computer science. suitable for medical reasoning along with its
advantages and disadvantages.
Artificial Intelligence contains a mechanism
behavior which has the capacity of 1. The Knowledge-Based solution
perception and learning, decision making This approach implements the process of
and intelligent reasoning. The feature human reasoning applied to a representation
intelligent reasoning of human being comes of knowledge [2]. The important component
with a rational approach. It creates tools in this approach is the knowledge base. The
which function better than humans as their Knowledge base consists of a set of related
mechanism are part of mathematics which sentences to the natural language. Therefore,
include logic, reasoning and engineering and it knows the things that an automated
are not prone to human errors [2]. decision system should know to generate an
optimal solution.
Benefits of AI in Medical Diagnosis
Health is an important aspect of everyone's In Figure 2, is an example of medical
life and central for our well- being. diagnosis of three types of Hepatitis: B,
Medicine is a domain where predictions and B+D and C. Rules are created for decision
immediate actions are very important [2]. making based on some laboratory tests. The
Automatic system is useful for two reasons. nodes of the graph consist of the knowledge
● Simple usual things which are time pieces or logical operators. The edge
consuming can be performed by a connects a pair of nodes.
machine and avoid overloading the
medical staff.
● Few medical condition are hard to be
detected by humans. An alert or a
suggestion can make a huge
difference between life and death.
different evolutionary types and several
severity levels. It would be useful to have
certain predictions on the patient's evolution
for the treatment. In such a scenario, due to
the complexity representing the patient's
premises in clear rules for Knowledge-based
system diagnosis is difficult. The solution
for this situation would be provided by
statistical methods. All the standardized
premises for as many patients as possible is
stored in a database to calculate the
probability of evolutionary type and severity
level [2]. The statistical approach provides a
probability of each possible output. Bayes
Figure 2: Structure of a knowledge based system for
Medical diagnosis [2].
theorem is used to implement the
probabilistic reasoning and the formula is as
Advantages shown below.
The knowledge representation is in the form
of natural language which is easy to create
and understand and therefore readily
accepted and used in various fields. It is
easy to implement for a knowledge base
with simple rules. The results are clear, uses The theorem describes the connection
logic and states whether the output is true or between two events Dk (which is the
false. A simple system represented by this diagnosis of many possible diseases) and S
mechanism always produces a correct result. (which stores patient's symptoms and
laboratory test results) [2].
Disadvantages
If the task to be solved has many data with Advantages
complex connections, the graph becomes too The solution offered by this mechanism is
complex. It becomes difficult to express the more accurate than Knowledge-based
logic to be used to decide on conclusions. mechanism.
Because of its complexity, the information
content could be damaged while getting Disadvantages
converted into rules. If the graph is too An important feature of this system is that it
complex, unexpected results can be does not provide a clear result, it just
encountered [2]. provides a probability. For example, it is not
sure based on this mechanism if a person
2. The Probabilistic Reasoning will develop a severity regarding a disease,
If the information system is too complex but it is more probable based on the system
then probabilistic approach is used for output [2].
decision making. For example, let's consider
a complex scenario where a patient is B. Augmented Reality for Future Medicine
diagnosed with hepatitis B. The problem in The concept of Augmented Reality is very
the condition arises with the evolution of the popular in various fields. The advancement
patient's medical condition which involves and popularity of mobile tech devices such
as tablet PC, smartphones and google In case of an emergency if a person has
glasses provides the community to develop a collapsed, getting medical care in a timely
lot of possible AR applications. Augmented manner is very critical for the patient's life.
Reality applications is gaining its popularity It takes time to call an emergency number
in Medical science field for education, and explain the situation and provide current
diagnosis, pre-planning and as well as in address to get help. In such a situation, an
operation room. Augmented Reality application can provide
the nearest hospital from its database. You
Benefits of using Augmented Reality for can even project the nearest hospitals on the
Medicine applications screen which provides the
AR is unique from Virtual Reality in two directions to the nearest hospital as well [7].
aspects; the user gets information as fast as
possible and as well as the information is in Proposed system for Rash Diagnosis using
eyesight, and user does not lose in touch AI and Augmented Reality
with reality. These feature enables AR in In this paper, we propose the use of
becoming a driving factor for many future Artificial Intelligence and Augmented
medical applications [7]. Reality in diagnosing a rash which will be
useful for a person living in a remote area
Augmented Reality acts as an assistant to who does not have access to local doctor for
surgeons in the OR. While operating in the inspection. Our approach attempts to mimic
surgery room time and precision are two a doctor’s medical assessment through
very important requirements. AR healthcare technology.
app can save lives by providing the right
information to the surgeons in a timely System Architecture Components
manner. An AR developed sunglasses
creates a 3-dimensional construction of a 1. AI
tumor which is visible by the surgeon ChatBOT AI is a computer program which
without any radiation exposure. The conducts a conversation via textual or
application could provide useful information auditory methods. It is designed to simulate
regarding the operation to the surgeon [7]. a human conversational pattern. In our
system, we ask few questions to the user for
Patients can visually see how a drug will diagnosis. Based on the answers given by
affect their body in 3D. Instead of reading the user it can reason and give diagnosis for
the long boring description of the medicine the rash.
on prescription bottle, the user can hover
their mobile device on the label and see its 2. Augmented Reality
details in 3D [7]. After the initial medical assessment of the
user based on chatBOT, further additional
AR technology can make the lives of both details are required from the user to
patients and nurses easier through AccuVein diagnose the rash. Using mobile Augmented
AR technology. The AccuVein uses AR to Reality, we collect additional data about the
act as scanner that projects over the skin. users’ rash through image processing. An
This projection shows the nurse where the AR equipped mobile device with camera is
veins are present in the patient's body. This hovered over the infected area. Using the
makes finding the vein on the first stick 3.5 image processing component of AR, we
times more likely [7]. compare the existing rash image with a
database of all rashes available to identify A. Rash Diagnosis: An Augmented Reality
the type of rash. Based on the answers System
collected from chatBOT along with rash This component basically needs to detect the
image processing through AR we can give type of the rash on the skin. Numerous
diagnosis to the user. algorithms use skin color as one of the basic
features for detecting and tracking human
3. Mobile Sensors being [2].
After we conclude the diagnosis based on ● Target: In our system, rash on the
the severity of the rash we collect the user's skin which needs to be diagnosed,
GPS location through the location sensors should be the target image. The
available on the device. Based on the current extracted rash area is used by the
user's location we provide the nearest tracker to recognize type of the rash.
doctor's location by communicating with ● Target collection: An archive storing
third party API’s. Additional data about the a collection of different types of rash
user's health is collected from other sensors that can be recognized by the tracker.
present in the users’ device and mailed to ● Tracker: The tracker analyzes the
the doctor. live camera image and detects the
targets stored in its associated target
collection. Multiple trackers can be
created, however, only one tracker
can be active for recognition at any
given time.
1. Tracking Methodology
This is the first phase of Augmented Reality
in which the images of the real scene are
taken by tracking cameras. There are three
main approaches; Marker based, Model
based and Motion based approach [13].
Figure 4 shows the tree structure of Mobile
Augmented Reality Taxonomy.
Steps
To incorporate and interact with as much of
the “real world” as possible, several steps
need to be executed:
1. Preprocessing: The target image that
should be tracked is analyzed.
Significant areas in the image, Feature Fig 6: The Wikitude Tracking Engine
Points, are extracted and stored. The
extraction algorithms are already
C. Machine Learning for Rash Diagnosis
explained in the previous chapter. This
Natural Language Processing(NLP), a part
step is executed only once per tracked
of Machine Learning, can translate and
image, and can run offline and
interpret human natural language input. The
asynchronously.
machine learns and uses its algorithms
through supervised and unsupervised
2. Feature Point Detection: In this step, the
learning. Supervised learning is to train the
current camera image is analyzed for
machine to convert the input into a required
unique points.
output. Unsupervised learning is to discover
new patterns in the data without any prior
information. The machine itself allocates a
function to the data through raw data
pattern. This is done via extracting the
hidden layers which are part of the data
processing layers in a neural network.
Methodology
There are four types of chatbot models and
generally this framework categorized in a
square model [12].
Retrieval-based models
This is an easier method in which responses
are stored in a repository and response is
picked based on the input and context. It
could be a rule-based expression match.
These systems don’t generate any new text, Fig 8. ChatBot Machine Learning model [15]
they just get picked from a fixed set of
responses [15]. Figure 8 explains how ChatBot works; the
first square block is contained, not complex
and can deliver both business and user
value. In the second square, questions are related data. HealthKit can effortlessly share
asked and the smart machine technology data between fitness and health sensors.
generates the response. Based on this There are two main data types;
response, Chatbot handles both the common Characteristics and samples.
questions and some unforeseen situations for Characteristics: User's birth date or blood
which there are no predefined responses. type.
The last block which is a smart machine can Sample: Represents data at a point in time
handle longer chats and it behaves more like which includes the user's height and weight,
a human. But generative response increases steps taken, temperature, pulse rate, etc.
complexity. [13] Workouts, which belong to the samples
category, are intended specifically for
D. Sensors representing runs, walks, rides, etc. Workout
Consider a scenario where a doctor in Delhi data might not be useful for doctor but
want to monitor the heart rate, and vitals characteristic and samples are the important
data such as blood pressure, body factors for any doctor to correct diagnosis.
temperature and respiratory rate of a patient
in Mumbai. This is done with the patient 3. Search Engine
having a wearable device such as apple Most search APIs are multi-index, multi-
watch that collects the data and sends to a type which can be applied across multiple
central system. The system then sends this types within an index, and across multiple
data to the doctor either through a mobile indices with support for the multi index
push notification or a live feed to his laptop syntax. There are various technologies
browser. available; Elasticsearch is one of the highly
scalable open-source full-text search and
1. Location Service Detection analytics engines. It allows you to store,
Using the standard location service in the search, and analyze big volumes of data
background, mobile application can use the quickly and have complex search features
location based tracking of user to detect the and requirements. There are some important
medical services nearby. The system applications which uses Elasticsearch to
provides location updates to background store large set of data and provides some
apps when the application is in the amazing techniques to search and
foreground, running in the background, or is autocomplete suggestions, collect logs, and
suspended. If the application is in suspended visualize data using Kibana dashboard [14].
mode, the system wakes up the application
and delivers update to location manager’s But recently, for standalone systems, there is
delegate and then goes back to the another simpler yet effective approach;
suspended state. While running in the Location based social networks (LBSNs) are
background, application should process new popular with the fast deployment of mobile
location data. Ideally, application should not networks. There are many LBSN providers
consume the location service continuously such as Facebook places, Google Earth,
as it degrades the battery life. Yelp. By exploiting the application
programming interfaces (APIs) provided by
2. Health Data Yelp, we will concentrate on Yelp and focus
The HealthKit framework for iOS was on sampling and estimating the places
announced at WWDC 2014 as a single place nearby patients. [10]. The POI search
where apps can store, share, and read health- knowledge services that we define are
wrappers around existing web services and interface for Augmented Reality will not
produces semantic web services [11]. In our look good and it will be inconvenient for the
case, we have programmed the logic for user to operate with a small screen [14].
lifting our different knowledge services that
may choose from different sources like VII. CONCLUSION
Yelp, Google Places since these domains are
constant. So, when the search is called on We believe that this system is an extremely
keywords such as “skin cancer doctor”, good implementation of AI and AR for skin
“acne specialist” etc. the JSON response of rash diagnosis. The proposed iOS
the list of nearby doctors is then parsed and application can be used by a person who
shown to user using appropriate logic. needs immediate help in skin rash diagnosis.
VI. CONCERNS The paper primarily focuses on skin diseases
like sun burns, which can lead to skin cancer
Even though our Proposed solution which and acne. The application will be able to
uses AI and Augmented Reality check the intensity of skin rash, and notify
technologies which theoretically proves to the person accordingly about the further
be convenient, efficient and exceeding steps needed to be taken.
human ability to perform calculations. The
technology should not be substituted for Further enhancements for this application
human judgment. The doctor is still the best will be developing the system for Android
standard for medical diagnosis [2]. users. The system collects the health
information from the health app of the
Privacy and security concerns arise in this phone. In future, the system will be
rash disease system. The patient might not enhanced to send these details to the doctor
want to discuss the course of the treatment for quicker medical help to the user on
and want to keep it confidentially. It needs arrival to the doctor’s clinic.
to be handled in a secure but simpler manner
[12]. REFERENCES
[1] Martin. (Oct 24, 2013). The wikitude
Since Android based mobile devices are Tracking Engine and Wikitude studio- The
Java based, the applications run on Java CTOs perspective [online]. Available:
virtual machine which consumes a lot of https://www.wikitude.com/the-wikitude-
computing resources on the loop of large tracking-engine-and-wikitude-studio-the-
trunk of data, which is needed by the image ctos-perspective/
processing application [13]. Compared to [2] Adriana Albu, Loredana Stanciu. (2015).
Android, iPhone has a better graphical Benefits of Using Artificial Intelligence in
processing unit (GPU) and supports C/C++ Medical Predictions [online].Available:
natively which means it can run more http://ieeexplore.ieee.org.libaccess.sjlibrary.
efficiently [13]. org/stamp/stamp.jsp?arnumber=7391610&ta
g=1
There are general issues with developing [3] Bhagya Hettige, Hansika Hewamalage,
mobile Augmented reality applications Chathuranga Rajapaksha, Nuwan
related to Navigation and tracking. It is hard Wajirasena, Akila Pemasiri, Indika Perera.
to track the location of the user inside the (2015). Evaluation of Feature-Based Object
building. Usability is another concern as the Identification for Augmented Reality
mobile device screen is very small, the user Applications on Mobile Devices[online].
Available:http://ieeexplore.ieee.org/docume https://books.google.com/books?id=zHO7B
nt/7399005/?reload=true&arnumber=73990 QAAQBAJ&pg=PA708&lpg=PA708&dq=
05 yelp+api+algorithm+ieee&source=bl&ots=
[4] Dimitris Chatzopoulos, Carlos Bermejo, GPnUWzFRaF&sig=c2ltn9L72SrK3f7OaX
Zhanpeng Huang, and Pan Hui. (26 April U3zYDC5yE&hl=en&sa=X&ved=0ahUKE
2017). Mobile Augmented Reality Survey: wjTkt6898_TAhUB6GMKHY8ADGIQ6AE
From Where We Are to Where We Go. IMzAD#v=onepage&q=yelp&f=false
[online]. [12] Julie Carmigniani. (Dec 2011).
Available:http://ieeexplore.ieee.org/stamp/st Augmented Reality Methods and algorithms
amp.jsp?arnumber=7912316 . for hearing Augmentation[online].
[5] Yao XIao-yang. (2011). Study on Available:
Development of Information Security and https://fau.digital.flvc.org/islandora/object/fa
Artificial Intelligence [online]. Available: u%3A3762/datastream/OBJ/view/Augmente
http://ieeexplore.ieee.org.libaccess.sjlibrary. d_reality_methods_and_algorithms_for_hea
org/stamp/stamp.jsp?arnumber=5750602 ring_augmentation.pdf
[6] Madhavi Dhingra; Manisha Jain; Rakesh [13] Fu Yu.Real Time Augmented Reality
Singh Jadon. (2016). Role of Artificial on Mobile Phone.[online].Available:
Intelligence in Enterprise Information http://web.eecs.umich.edu/~silvio/teaching/
Security: A Review [online]. Available: EECS598_2010/final_report/Fisher.pdf
http://ieeexplore.ieee.org.libaccess.sjlibrary. [14] Stan Kurkovsky, Ranjana Koshy,
org/stamp/stamp.jsp?arnumber=7913142 Vivian Novak, Peter Szul. (2012). Current
[7] The Medical Futurist. (2017). Issues in Handheld Augmented Reality
Augmented Reality in Healthcare Will Be [online].
Revolutionary [online]. Available:http://ieeexplore.ieee.org.libacces
Available:http://medicalfuturist.com/augme s.sjlibrary.org/stamp/stamp.jsp?arnumber=6
nted-reality-in-healthcare-will-be- 285844
revolutionary/ [15] Henk Pelk. (Feb 16).Machine Learning,
[8] Eric Marchand, Hideaki Uchiyama, Neural networks and algorithms [online].
Fabien Spindler. (18 December 2015).Pose Available:
Estimation for Augmented Reality: A https://chatbotsmagazine.com/machine-
Hands-On Survey [online].Available: learning-neural-networks-and-algorithms-
https://hal.inria.fr/hal-01246370/file/survey- 5c0711eb8f9a
ieee-v2.pdf [16] Osian Haines. (May 13 2016). An
[9] (April 6, 2016). Deep Learning for Introduction to Simultaneous Localisation
Chatbots Part 1- Introduction [online]. and Mapping[online]. Available:
Available: https://www.kudan.eu/kudan-news/an-
http://www.wildml.com/2016/04/deep- introduction-to-slam/
learning-for-chatbots-part-1-introduction/
[10] Yanhua Li, Moritz Steiner,Jie Bao.(19
May 2014).Region sampling and estimation
of geosocial data with dynamic range
calibration [online]. Available:
http://ieeexplore.ieee.org/document/6816726
[11] Elena Simperl, Philipp Cimiano, Axel
Polleres, Oscar Corcho, Valentina Presutti
[online]. (2012). Available: