www.ijcrt.

org © 2023 IJCRT | Volume 11, Issue 6 June 2023 | ISSN: 2320-2882

ARTIFICIAL INTELLIGENCE IN MEDICAL

DIAGNOSIS
(Use of Algorithms & Software to Approximation Human Cognition in the Analysis of Complex Medical Data)

Abhilasha Rajput1, Prof. (Dr.)Pushpneel Verma2, Ajay Singh3

1Student, 2Professor
& Research Guide, 3Assistant Professor & Head of Department
Computer Science & Engineering Department, Bhagwant Institute of Technology (BIT), Muzaffarnagar, U.P., India

ABSTRACT - In the field of medicine, predictions are information is preceded. Artificial Intelligence has the
of great significance. In recent times, intelligent systems capability to predict results and diagnose illness at a
play an important role. The techniques of the user rate which is higher than most medical professionals.
interface are built on the ever-increasing and and A healthcare decision-making interface can make
progressing areas such as Artificial Intelligence (AI) and use of technology and enhance the diagnosis process.
Artificial Neural Networks (ANN). The mechanism of
An AI-powered medical diagnosis system can rapidly
artificial intelligence provides substantial assistance in
healthcare. Considering these optimistic aspects, a store an immense amount of data and can make
combination of accuracy of mathematics and the intricate connections between them. Medical diagnosis
potentiality of automation results in a robust system. In applications incorporate different mechanisms from
this paper, our aim is to develop a healthcare interface the large domain of artificial intelligence, highlighting
web application that will aid the user and enable them to the vast amount of benefits they have been able to
issue their symptoms of common diseases or issues faced achieve to the field of medical decision making and
by the user to generate a prognosis. The primary also these mechanisms come with their own
objective is to develop a machine learning system that drawbacks which has been discussed as it is essential
utilizes AI and deep learning to help people to keep a to emphasize them to decide a favorable AI method
check on their health with the ease of a web application. for a specific task. Some of these features (pros and
Artificial intelligence (AI) aims to bring a revolutionary cons) are present in the literature survey of the domain
change to the healthcare sector, reinforced by the which act as a strong evidence by the decisional
increasing availability of healthcare from hospitals systems presented. Some others have been noticed
which acts as a catalyst to the rapid progress of once these systems were developed [5].
analytics techniques.
This paper proposes a system that is used for the
Key Words: Prediction, User Interaction, Artificial prognosis of the users experiencing symptoms which
Intelligence, Artificial Neural Networks, Prognosis, may cause discomfort to the user but can be quickly
Machine Learning. analyzed with the help of a self-service web
application to diagnose the issue and if severe, contact
1. INTRODUCTION a medical professional. The system is trained and
tested using machine learning based on a dataset of
Artificial intelligence is a cohesion of multiple symptoms faced by the users from the UCI Machine
technologies grouped together. Medical diagnostic Learning Repository [6].
applications utilize the AI approach to diagnose disease. The paper is constructed as follows: Section II
These technologies have a vast variety of immediate discusses the survey which was conducted, Section III
relevance to the field of healthcare and nutrition, but the and Section IV talks about the proposed system and
specific ways in which they process may vary widely. Section V concludes the paper.
Today, the swift evolution and innovation of applied
science have altered the perspective and the way

IJCRT2306926 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org h786

www.ijcrt.org © 2023 IJCRT | Volume 11, Issue 6 June 2023 | ISSN: 2320-2882
2. LITERATURE SURVEY human brain. It can be used to cluster unlabeled data
and if labeled can help to classify them. The basic
The purpose of carrying out a Literature Survey is to structure of ANN consists of four parts which can be
demonstrate and develop our familiarity with the represented in the figure 2 below as we can see it is
existing work relevant to the focus of our study. segregated into the Input, Middle and Output layer as
well as sensor nodes which help in providing the
data.[1]
Prediction Techniques
Machine learning algorithms are used to analyze data
again and again to produce the most effective results. It
can be used for the analysis of medical data and it is
helpful in medical diagnosis for sensing different
diagnostic problems.
A useful method called the decision tree is used to
formulate an expression which is used and incorporated
in such mappings and also consists of test which may
also be attributed to multiple nodes which are linked to
two sub-tress or more and the leaf-nodes which is the
decision is linked with the help of a labeled class.
The decision trees though being very accurate and also Figure -1. Basic Structure of ANN
efficient often suffer from excessive complexity. In the The gathering of information is done using the
case of an unsupervised evaluation the function can sensor nodes.
become inappropriate easily which in result will result in
The actual processing starts with the input layer. The
a bad solution [7]. Over fitting is one such term where
input layer applies an activation function over the
the novice decision tree creators can unintentionally
input nodes. The input layer data is multiplied with
create over-complex trees which doesn’t generalize the
weights before forwarding it to the next layer and so
data well and has problem in doing so. Population size is
on as shown in the figure. The hidden layer does the
also quite important: if the solution is very large then it
work of processing and is the main reason for the
will improve very slowly through the generations.
acceleration of processing work. The output layer
Bayes Theorem utilizes vital information which is gives the final result.
already known about the predictive value of an
In the proposed paper, the author tried to apply a
observation which is relative to the given outcome to
heart diagnostic system to detect a synergy in the
modify the probability of a known particular outcome
cardiac muscles using neural networks and fuzzy
[8]. Naïve Bayes classifier is probabilistic in nature
logic [2]. In the proposed paper the author built a
which is primarily based on applying the Bayes theorem
self-diagnosis system for detecting cervical cancer
which have strong independence assumptions between
with the help of artificial neural network [6]. In this
the parameters. Bayes theorem can be deployed used for
paper the author tried an experiment of testing an
prediction of the diseases as a classifier’s decision rule.
ANN based diagnostic model. The author described
The probability of the disease according to its number of
about the architecture of ANN being used and came
symptoms can be predicted using Bayes algorithm [9].
to a conclusion of the accuracy and reliability of a
One of the disadvantages that is considered as a choice
typical architecture after performing several
is that Bayes theorem expresses the probability based on
experiments on the model [1].
three other probabilities.
Neural networks are algorithms designed based on the

IJCRT2306926 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org h787

www.ijcrt.org © 2023 IJCRT | Volume 11, Issue 6 June 2023 | ISSN: 2320-2882
In the following paper the author did a review of the Cura No Yes
technologies being used in the field of medicare. The
Proposed System Yes No
author also described the implementation of numerous
AI technologies in the field of medical sciences and tried Table -1: Trait of the existing self-diagnostic systems
to conclude the accuracy of techniques being used in the and the proposed system.
medical diagnosis [3].
In reference to the table, the following observations
A. Medical Self-Diagnosis Systems.
can be concluded:
As we all know the major areas in medical fields where
Most of the systems are not free.
decisions play a vital role especially in the medical
diagnostic analysis and process as there are several Most of the systems are attached to a hospital.
studies on medical diagnostics. To determine or interpret Features are restricted to the users which results in a
the illness based on the symptom/s experienced by an lack of flexibility and individuality. An easy to use
individual is known as medical diagnosis. user interface is much appreciated that is freely
A user-interface application which may be in any form available.
such as a mobile application or web application aims
just to do that: the individual enters the symptoms that
they’re experiencing into the app and a prognosis is B. Contemplated Model.
generated which is made possible with the Medical care is vital for all humans. This paper
implementation of intelligent machine learning attempts to fill these gaps and proposes a system that
algorithms some of which are incorporated in the is free, general, i.e., not attached to a specific hospital.
proposed system. A simple automated system that diagnosed symptoms.
A chatbot called as Babylon Health’s symptom checker Also, assimilate several classifier techniques to aim
is one such application where the user can interact with for better accuracy of the system.
the system and generate a prognosis on the input given METHODOLOGY
by the user, the system follows a general QnA (Question
Front End
and Answer) format to collect information from the user.
The system proposed here is deployed as a Web
Also, there exist several systems such as Mayo Clinic
Application which is enabled with the help of web
(https://www.mayoclinic.org/) and Cura
development tools like HTML, CSS and JavaScript in
(http://cura.healthcare/en/), usually these systems are
the front-end which is incorporated with a back-end
affiliated to certain hospitals and only available for the
that uses a machine learning model for symptom
patients of these hospitals. Also, many of these systems
diagnosis. The web application greets the user with a
are not free [7].
minimalistic yet simple user interface that can be
Table 1 shows a list of medical self-diagnostic systems. easily navigated by users of all age groups since the
The table presents the system name, whether it is free or web application is clutter-free as it just takes user
not, whether it is attached to a specific hospital. The last symptoms and provides a diagnosis to the user. While
row shows the features of the proposed system as using the web application the users should provide
compared to the existing ones. accurate information to get the correct diagnosis. The
web application is integrated with text fields in which
Attached to a the user can increase or decrease depending on the
Name Free
Hospital symptoms experienced. Designing the graphical user
Mayo Clinic No Yes interface (GUI) was preferred over a chatbot as having
an easy to use interface is the priority as mentioned
Babylon No Yes earlier.

IJCRT2306926 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org h788

www.ijcrt.org © 2023 IJCRT | Volume 11, Issue 6 June 2023 | ISSN: 2320-2882

A simple probabilistic classifier known as the Naïve

Bayes Classifier is based on applying Bayes theorem
with strong independence between the features. They
are one of the simplest Bayesian network models but
can achieve greater accuracy by combining them with
Kernel Density Estimation. Bayes theorem calculates
the posterior probability, P(c|x), from P(c), P(x), and
P(x|c). Class conditional independence is one of the
assumptions made by Naïve Bayes classifier which
assumes that the effect of the value of a predictor (x)
on a given class (c) is independent of the values of
other predictors.

Figure -2. Block architecture of the system.

The web application is designed using basic web

programming languages such as HTML, CSS and
JavaScript. The middleware used for connection is
Flask. Flask is a python framework which provides with
libraries and tools for developing web application. In the
system proposed the data from the user interface is
transferred to the backend using flask which helps in Figure -3. Naïve Bayes Theorem
making API calls to the model for prediction and the For applying this algorithm on python we can use
predicted result is fetched and displayed on the GUI. multinomial naive bays classifier from sklearn python
library. The following algorithm can be applied using
multinomial NB classifier on the train-test split format
Request is used for making API calls when triggered by of dataset where 80% is recommended for training and
the interface, jsonify is used for fetching the data the remaining for testing the accuracy of the ML
collected from the UI and render template is used for algorithm trained.
displaying results on the web application.
Another supervised learning algorithm that can be
applied is Artificial Neural Network (ANN) classifier
Back End which is based on neural networks. Neural networks
are algorithms designed based on the human brain. It
The backend of the application is a supervised machine
learning model. can be used to cluster unlabeled data and if labeled can
help to classify them. As the dataset used is in a
labeled form, we can use ANN as a classifier on it.
Based on the literature survey and the dataset considered The basic structure of ANN consists of four parts:
the following two classifiers can be used: sensor nodes, Input layer, Middle layer, and Output
layer as shown in figure 2.[1]
The gathering of information is done using the sensor
1. Naive Bayes Classifier
nodes. The actual processing starts with the input
2. ANN Classifier layer.

IJCRT2306926 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org h789

www.ijcrt.org © 2023 IJCRT | Volume 11, Issue 6 June 2023 | ISSN: 2320-2882
The input layer applies an activation function over the classifier model’s accuracy to be 91.4% and that of
input nodes. The input layer data is multiplied with Naive Bayes to be 86.3%. Based on the accuracy
weights before forwarding it to the next layer and so on scores ANN classifier provides more accurate
as shown in the figure. The hidden layer does the work predictions for the selected dataset. This system would
of processing and is the main reason for the acceleration be beneficial for the users to take pre- diagnosis so
of processing work. The output layer gives the final that it can gauge the severity of the symptoms being
result. In our project, the dataset is provided to the input experienced. By applying this system as a web
layer where symptoms become the input nodes, and application, the hassle of downloading the app and the
prognosis becomes the nodes in the output layer. The barrier of the technology used in the user’s devices are
hidden layer nodes are adjusted based on the input and eliminated. In future the data fetched by the user
output nodes. (using only the symptoms and predictions excluding
the user’s personal information) could be updated in
the dataset and the model could be retrained at regular
intervals to get better accuracy of the model.

4. REFERENCES
1. Shanghui Yin, Renzhi Xing, Xiangqi Liu, Yinhui
Yi, Kai Zheng, Xin Huang- “Model Checking an
Artificial Neural Networks System in Medical
Diagnosis”, 2018 9th International Conference on
Information Technology in Medicine and Education
(ITME),IEEE,2018.

Figure -4. Functioning of a Supervised Neural 2. Yasue Mitsukura, Kayoko Miyata, Kensuke
Network. Mitsukura,Minoru Fukumi, Nono Akamatsu –
For creating the model dense and sequential are the
functions of the keras library used.
3. “Intelligent Medical Diagnosis System Using the
Sequential specifies that the model being formed is in Fuzzy and Neural Networks*”, IEEE ,0-7803- 8376-
sequential format and the output of the layer which have 1/04,2004.
been added is the input to the next layer which is going
4. Luteshna Bishnoi, Dr.Shailendra Narayan Singh-
to be added. As a supervised model is being applied the
actual result is compared with the desired one to give the
error signals which in succession are used to adjust the 5. “Artificial Intelligence Techniques Used In
weights as shown in the figure 4. In this way after Medical Sciences: A Review ”,IEEE,978-1-5386-
several epochs the model is finally trained which can be 1719-9/18,2018.
used for prediction.

6. Dr. Sivanandam, S.N. and Dr.. Deepa,

4. CONCLUSION S.N., 2011, Principles of Soft Computing, Wiley,
Thus, this paper represents a supervised machine India, 619p
learning algorithm driven system for simple medical
diagnosis of the user. The result of applying the
7. Adriana ALBU1, Loredana STANCIU2- “Benefits
algorithms on the dataset used suggests that the ANN
of Using Artificial Intelligence in Medical

IJCRT2306926 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org h790

www.ijcrt.org © 2023 IJCRT | Volume 11, Issue 6 June 2023 | ISSN: 2320-2882
Predictions”, IEEE, 978-1-4673-7545-0/15

8. The 5th IEEE International Conference on E- Health

and Bioengineering - EHB 2015

9. Malikah Aljurayfani, Sundus Alghernas, Amal

Shargabi- “Medical Self-Diagnostic System Using
Artificial Neural Networks”, IEEE, 978-1- 5386-8125-
1/19, 2019.

10.Vili Podgorelec- “Self-Adapting Evolutionary

Decision Support Model”, IEEE, 0-7803-5662- 4199

11. “A general purpose shell for research

assessment of bayesian knowledge bases supporting
medical diagnostic software systems”, IEEE, CH2845-
6/90/0000/0267

12. Himdeep Bohra, Amol Arora, Piyush

Gaikwad, Rushabh Bhand, Manisha R Patil, “Health
Prediction and Medical Diagnosis using Naive Baye

IJCRT2306926 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org h791