TOPIC: Kidney Stone Detection using Image Processing

A Major Project Report was submitted in partial fulfilment of the Bachelor of

Technology degree requirement.in

Computer Science and Engineering

By

NISHKARSH SHARMA (181258)

UNDER THE SUPERVISION OF

DR. SURJEET SINGH

Department of Computer Science & Engineering and Information

Technology
Jaypee University of Information Technology, Waknaghat,
173234, Himachal Pradesh, IND
 Certificate

Candidate’s Declaration

I hereby declare that the work presented in this report entitled “Kidney Stone
Detection using Image Processing” in partial fulfillment of the requirements
for the award of the degree of Bachelor of Technology in Computer Science
and Engineering/Information Technology submitted in the department of
ComputerScience & Engineering and Information Technology, Jaypee
University ofInformation Technology Waknaghat is an authentic record of my
own work carried out over a period from August 2021 to December 2021 under
the supervision of Dr. Surjeet Singh Assistant Professor(SG) Department of
Computer Science and Engineering.
The matter embodied in the report has not been submitted for the award of any
other degree or diploma.

NISHKARSH SHARMA
Nishkarsh Sharma, 181258

This is to certify that the above statement made by the candidate is true to the best
of my knowledge.

(Supervisor Signature)
Dr. Surjeet Singh
Assistant Professor(SG)
Computer Science and Engineering
Dated:

I
 AKCNOWLEDGEMENT

To begin, I want to express my heartfelt gratitude to almighty God for His

heavenly grace, which has enabled us to successfully complete the project work.

Supervisor Dr. SURJEET SINGH, Assistant Professor (SG), Department of CSE

Jaypee University of Information Technology, Waknaghat, deserves my deepest
gratitude. His never-ending patience, intellectual direction, constant
encouragement, constant and energetic supervision, constructive criticism,
helpful suggestions, and reading numerous poor versions and revising them at
every level allowed this project to be completed.

I'd like to thank Dr. SURJEET SINGH, Department of CSE, for his invaluable
assistance in completing my project.
I'd also like to express my gratitude to everyone who has assisted me in making
this project a success, whether directly or indirectly. In this unique situation, I'd
want to express my gratitude to the different staff members, both teaching and
non-teaching, who have provided me with valuable assistance and assisted my
project.

Finally, I must express my gratitude for my parents' unwavering support and

patience.

Nishkarsh Sharma

II
 ABSTRACT

Kidney stones can be a huge problem on some days, and if they aren't caught early enough, they
can create problems, necessitating surgery in addition to the tests that are required to find the stone.
The ability to notice a stone that is clearly visible is due to image processing, since processing the
picture has the tendency to encourage correct and automated outcomes in how to discover a stone.
Due to the presence of noise in the images, the distinction of kidney stones is imprecise. Kidney
stones have become more common in recent years as a result of a variety of causes. Using human
and operational tests to draw conclusions for large datasets is tough. It's tough to pinpoint the exact
place where a kidney stone during surgery. Kidney stone illness is one of the most serious diseases
on the planet. An automated kidney stone detection system is built using digital imaging and data
processing approaches. On CT scans and MRIs, there is a lot of noise, which adds to low accuracy.
Neural network-based artificial intelligence technologies have yielded considerable advances.
Artificial intelligence, such as neural networks, has proven to be quite useful in this field. As a
consequence, Digital Image Processing is being employed in this project and furthermore we have
tracked the accuracy and comparison has been made with the existing approach carried in the
healthcare industry.

III
 Table of Content

Sr. No. Particulars Page

Certificate Declaration I
Acknowledgement II
Abstract III
1 Introduction of Major Project 2-5

1.1 Introduction 2
1.2 Objectives of Project 3
1.3 Technical Requirements 4
1.4 Deliverables of Project 5
2 Literature Survey 6-11

2.1 Research on Existing Approaches and Recent Researches 6-7

2.2 Feasibility study on Project 8
2.3 Requirements on Major Project 9
2.4 Objective of Major Project 10
2.5 Use Case Diagram of Major Project 11
3 System Development/ Implementation 12-43
3.1 Dataset Used in Major Project 12
3.2 Data Set Features 13
3.3 Design of Problem Statement 14
3.4 Algorithm/Pseudocode of Project with Screen Shots of Project 15-43
4 Performance Analysis 44-50
4.1 Discussion on Result Achieved 44-49
4.2 Project Outcome 50
5 Conclusion 51-55
5.1 Conclusion 51
5.2 Application of Queue Management System 52
5.3 Limitations of Queuing Model 53
5.4 Contributions 54
5.5 Future Work 55
References 56-58
1
 CHAPTER 01

1. Introduction

The doctor often utilizes the hand approach to find the stone in the X-ray picture, but our method
is totally automatic, which saves time and reduces the risks of making a mistake. This project
describes a method for detecting kidney stones that includes image processing processes. The basic
function of the kidney is to maintain electrolyte balance in the blood. A bean-shaped organ, the
kidney can be found on either side of the spine. Kidney stones are becoming more widespread
around the world, and most people with concretion illness aren't aware of it because it destroys
organs gradually before symptoms appear. There were 16000 papers on detecting kidney stones
using various types of filters in Google Scholar and IEEE Xplore in the previous 5 years. To locate
photos with poor contrast and speckle noise, digital image processing is performed. An automated
kidney stone detection system is built using digital imaging and data processing approaches. There
is a great deal of noise on CT scans and MRIs, which beneficial to low accuracy. Artificial
intelligence methods based on neural networks have shown significant results. Artificial
intelligence, such as neural networks, has shown to be quite helpful in this area. As a result, in this
project, the Digital Image Processing is being used on ultrasound sound images. Image - processing
applications include image sharpening and restoration, colour processing, and pattern recognition.
Low contrast and speckle noise in ultrasound images are drawbacks. The detection of kidney
stones is a difficult task. Speckle noise is a natural characteristic of medical ultrasound imaging,
and it reduces image resolution and contrast, lowering the diagnostic usefulness of the imaging
modality. The kidney stone is found using ultrasound images and then removed with a surgical
process that includes breaking up the stone into smaller bits that can be passed down the urinary
tract. Because kidney stones have low contrast and speckle noise, detecting them is a difficult task.
This problem is solved by employing the appropriate imaging techniques and filters. Speckle noise
is common in ultrasound images, which cannot be eliminated with standard filters. As a result, the
median filtering algorithm is proposed, which reduces speckle noise. A median filter is used to
eliminate noise and detect the stone region in the pre-processed image.

2
2. Objective of the Major Project

Kidney stones is difficult to track early because of hidden symptoms. There is a huge burden on
patients and the resilience system due to delays in diagnosis. The problem of kidney stone detection
has encouraged researchers to expand their research in order to track and detect this problem at an
early stage for the people with kidney stones and medical treatment should be given on time to avoid
any further complications. We have used Ultrasound images to detect the kidney stones and applied
techniques to predict accurate results. The reason why we collected the dataset of ultrasound images
rather than CT scan images(which is widely used in the healthcare domain to predict results) is that
we want to create an user friendly project which can be further implemented and can be used in the
daily life by the healthcare system to generate and track the accuracy of the results within the short
spare of time and further creating an ease to both healthcare system and to the common people with
the disease.

3
 3. Technical Requirements (Software)
(1) MATLAB
(a) For Implementation
(2) MS-EXCEL
(a) For Comparing Accuracy and Senstivity
4. Technical Requirements (Hardware)

As we will be working with

MATLAB

The Minimal System Specifications are:

● Memory: 4 GB
● Free Storage: 2 GB
● Screen Resolution: 1200 x 800
● OS: Windows 7/8/8.1/10 (64-bit)

The suggested system requirements are

● Memory: 8 GB RAM
● Free Storage: 4 GB (SSD Recommended)
● Screen Resolution: 1920 x 800
● OS: Windows 10 (64-bit)
● CPU: Intel Core i5-8400 3.0 GHz or better.

4
 5. Deliverables of the Major Project

The project will predict whether the person is or is not having Kidney stones inside his/her kidney. Using
the given data set we will analyze data using Image Processing Techniques (Using of Negative Image
Filter, High-Low Contrast Filter, Edge Detecting filters and so on). Our goal is to attain Accuracy and
sensitivity atleast more than the current methodologies existing in the healthcare industry so that this
model can further be used and can be implemented in the real world and which makes ease to both the
patient(s) and medical practitioner(s). We would also be looking forward to compare our approaches
taken to finish this project and will try to obtain and predict which approach gives us the better results
output based on numeric data

5
 CHAPTER 02:

LITERATURE SURVEY

1. Existing approaches and recent research on the similar kind of project:

No. References' titles Algorithm/Techniques Significance Measuring

1 A method for Threshold Technique: CT scan image Precision
segmenting kidney In this we convert a picture segmentation pre- Accuracy
stones in CT scan from colour or grayscale to processing Sensitivity
pictures using binary by changing pixel of
image pre- image to make analysis easy.
processing
2 SVM and KNN are K Nearest Neighbour & Image pixel count Recall
used to analyse and Support Vector Machine and stone detection Precision
identify kidney in kidney pictures
stones.
3 A study of a K Nearest Neighbour: Various processing Accuracy
systemic strategy to It finds the most common class techniques are
utilise ultrasound among the k-closest instances used to segment
images to detect to classify unknown data items. data.
renal calculi using
several analysis
methods.
4 Introduction using Gray Level Co-occurrence The image is Precision
GLCM and pixel Matrix Future Extraction enhanced and the Accuracy
intensity matrix low parts are Sensitivity
parameters to filtered.
differentiate
staghorn and
struvite kidney
stones
5 Detection of kidney Segmentation The quantity and Precision
stones in computed placement of the Accuracy
tomography images stones can be Sensitivity
determined with
the use of
segmentation.
6 Kidney Stone Image Processing Techniques Differentiation of Accuracy &
Detection in filters using Sensitivity
Ultrasound Images ultrasound images
Using the Median
Filter vs. the Rank
Filter

Table 1.
6
We have researched the above techniques/approaches to implement our project after analyzing
existing techniques/approach in the healthcare and technology domain. On the basis of our research,
we have created our own new approach taking references from our research part and successfully
implemented our project and later we have compared it with the existing technology working on
similar kind of projects.

7
2. Feasibility Study on Major Project:

This project predicts the presence of kidney stones inside an individual’s kidney using ultrasound
pictures. One of the imaging approaches for identifying kidney problems is ultrasound imaging, which
can include changes in terms of size and placement as well as swelling of the limb. Other kidney
abnormalities include the formation of stones, cysts, urine blockage, congenital anomalies, and
cancerous cells. It is critical to detect the real and accurate position of a kidney stone during surgical
procedures because the contrast is low and speckle noise is present , detecting them using ultrasound
imaging is a difficult process. This problem may be solved by using the right image processing
techniques.

8
3. Requirements on Major Project:

1. Functional Requirements

MATLAB

➢ Microsoft Windows 10

➢ Minimal: Any Intel or AMD x86-64 processor.

➢ Suggested: Any Intel or AMD x86-64 processor with four logical cores and
AVX2 instruction set support

➢ Minimal: 3 GB of HDD space for MATLAB only, 5-8 GB for a typical

installation.

➢ Suggested: An SSD is recommended.

MS-Excel

➢ Microsoft Windows 10

➢ The configuration required for the MATLAB would be enough to run and
process MS-office.

2. Non-Functional Requirements

ATLEAST: 4 GB memory

➢ Intel 8th generation higher CPU’s

➢ Nvidia 1650 min GPU

9
4. Objective of Major Project:

The goal of this project is to detect kidney stones using a dataset of ultrasound pictures (which includes
50% of images with kidney stones and 50% of images without kidney stones). This research will be
undertaken using digital image processing techniques such as applying various filters to the dataset,
and it will also be based on research acquired from other sources. This project has been separated into
two sections, with Part 1 being the Median Filter and Part 2 being the Edge Detection Filter. To
simulate and develop the code, we utilized MATLAB. We will compare the produced pictures on
MATLAB to track the accuracy and sensitivity obtained after successfully acquiring the results from
both filters on the dataset image. In addition, we will compare the findings in MS-Excel.

Once the code has been processed, the output picture will be seen in the command prompt. To acquire
the result, we will carry out the experiment again with distinct kidney ultrasound pictures and use the
algorithm to compute the rate of identification.
The input photos, which are independent variables, are ultrasound scans of kidney stones. Accuracy
and sensitivity will be the output variables.

10
5. Use Case Diagram of the Major Project

Input •Input Raw

Images from
Images Dataset

Image •Applying
Image
Filters Filters

•Pre-
Image processing
and Post-
Processing Processing
of Image

•Comparison
of filters and
Comparison accuracy on
images

•Based on
Analysis,
Results Predicting
Results

Fig 1. Use case Diagram

11
 CHAPTER 03

System Development

3. Implementation of the Major Project

3.1 Dataset Used in the Major Project

This database contains a large number of biomedical ultrasound images from various hospitals of
USA. There are total 228 Ultrasound images with 50% images that can have a possibility of kidney
stones and rest 50% normal images without kidney stones. Following are the few images from the
dataset which we will use.

Fig 2. Images from Database

12
3.2 Date Set Features

3.2.1. Types of Data Set

There is total 228 raw Ultrasound images taken from Kaggle as well as few from some Hospital’s

website to do this project. Every raw image before processing has nil signal to noise ratio until

some filter has been applied on it. All of the dataset is somewhat greyscale images and all of

them have properties of ultrasound images. The dataset used in this project have been collected

from different sources both available online and offline from hospitals/clinics.

13
3.3 Design of Problem Statement

More than half a million patients visit emergency departments each year with kidney stone issues. A
kidney stone affects one out of every 10 people at some point in their lives. In the United States, kidney
stone prevalence increased from 3.8 percent in the late 1970s to 8.8 percent in the late 2000s. In the years
2013–2014, 10% of people had kidney stones. Men have an 11 percent chance of developing kidney
stones, while women have a 9 percent chance. High blood pressure, diabetes, and obesity are all
conditions that increase the risk of kidney stones. Kidney stone disease is a serious health problem that
affects people all over the world. Stone disorders go undiagnosed in the early stages, causing harm to
the kidneys as they progress. Kidney failure affects the majority of individuals owing to diabetes,
hypertension, glomerulonephritis, and other factors. Because kidney disease can be dangerous, it's best
to get a diagnosis as soon as possible. Ultrasound (US) is one of the most often utilized noninvasive,
low-cost imaging modalities for evaluating renal diseases. Kidney failure is a potentially fatal condition.
Therefore, early diagnosis of kidney stones is crucial. To assure the accuracy of medical procedures, it
is necessary to precisely detect kidney stones. The ultrasound images of the kidney contain speckle noise
and weak contrast, making kidney abnormalities difficult to recognize. As a result, doctors may find it
tough and confusing to detect small kidney stones and their nature.

14
3.4 Algorithm / Pseudocode of the Project Problem with the screenshots of various
stages of the project

1. Creating A Graphical User Interface (GUI Model) on MATLAB

App Designer is an interactive environment that combines the two core aspects of app development:
visual component layout and behaviour scripting. It enables you to switch between visual design on
the canvas and coding in the MATLAB editor easily. With the help of App Designer we have created
an app environment to run our project. Following images will show the step by step procedure how we
have created an app environment for our project.

Fig 3.

STEP 1
Created a GUI where we can implement our project

15
 Fig 4.

STEP 2

Created a separate toolbox where we can draw filter buttons and other required buttons.

16
 Fig 5.

STEP 3

Created buttons using designer toolkit in the tools section that we have created in the previous step

17
 Fig 6.

STEP 4

To create an user friendly environment we have further created 4 buttons where we can also take input
from the user and accordingly the prediction will be made

18
 2. Explanations of filters used in the project
A. MEDIAN FILTER

Median filtering is a nonlinear method for minimising impulsive noise, often known as salt-and-pepper
noise. It may also be used to keep an image's edges while decreasing random noise. A random bit
mistake in a communication line can cause impulsive or salt-and-pepper noise. The median intensity
value of the pixels within the window becomes the output intensity of the pixel being processed in a
median filter, which slides along the picture. Assume that the pixel values within a window are 5,6, 55,
10, and 15, and that the pixel being processed is 55. The median filter produces a value of 10 at the
current pixel position, which is the median of the five values. The median filter is a non-linear digital
noise reduction approach for images and data. Noise reduction is a common pre-processing procedure
used to improve the outcomes of subsequent processing (for example, edge detection on an image).
Median filtering is commonly employed in digital image processing because it retains edges while
reducing noise under certain conditions (though see the explanation below), and it also has uses in
signal processing. The median filter's fundamental principle is to go over the signal entry by entry,
replacing each one with the median of the entries next to it. The "window" is a pattern of neighbours
that glides over the entire signal, entry by entry. The simplest evident window for one-dimensional
signals is merely the first few preceding and following entries, but the window for two-dimensional
(or higher-dimensional) data must encompass all entries within a certain radius or ellipsoidal region
(i.e. the median filter is not a separable filter).

Fig 7.
Median Filtered Images

19
 B. Negative Filter

The original picture's intensity I is replaced with 'i-1,' resulting in the darkest pixels becoming the
brightest and the brightest pixels becoming the darkest. By removing each pixel from the greatest
intensity value, an image negative is created.
An 8-bit grayscale picture's highest intensity value, for instance, is 255, so each pixel is subtracted
from that value to get the final/result image. An image's negative is one in which the lightest portions
appear to be the darkest and the darkest areas appear to be the lightest.
In a grey scale picture, the change in pixel intensity values from highest to lowest and lowest to highest
causes the appearance shift from lightest to darkest and darkest to lightest..
Different colours are represented as negatives of different colours in a colour picture based on their
intensity values.

Fig 8.

Negate filter images

20
C. Prewitt Edge Detection Filter

The Prewitt operator extensively used in edge detection approaches in processing of images. It is a
separate differentiation filter that calculates a rough estimate of the gradient of the picture intensity
function. At each point in the image, the filter returns either the relevant gradient vector or the norm
of this vector. It is a gradient-based operator, as the name implies. It is one of the most effective
methods for determining an image's orientation and magnitude. It computes the gradient
approximation of the image intensity function for image edge detection. The Prewitt operator
provides either the normal to a vector or the associated gradient vector at the pixels of an image. It
calculates approximations of the derivatives using two 3 x 3 kernels or masks convolved with the
input picture — one for horizontal changes and one for vertical changes. When we apply this mask
to an image, the vertical edges become more visible. It simply calculates the difference of pixel
intensities in an edge area, similar to a first order derivate. Because the centre column is 0, it does not
contain the image's original values, instead calculating the difference between the right and left pixel
values around that edge. The edge intensity is increased, and the picture is boosted in comparison to
the original.

Fig 9.
Prewitt filter image

21
D. Canny Edge Detection Filter

The Canny edge detector is a multi-stage edge detection operator that detects a wide variety of edges
in images. It was created in 1986 by John F. Canny. Canny edge detection is a method for obtaining
structural information from a variety of visual objects while drastically lowering the amount of data
to be analyzed. It's been used in a variety of computer vision systems. Canny discovered that the
criteria for using edge detection on a variety of vision systems are quite similar. As a result, an edge
detection system that meets these requirements can be used in a variety of scenarios. The Canny filter
is a multi-stage edge detector. It uses a filter based on the derivative of a Gaussian to determine the
gradient intensity. The Gaussian filter reduces the influence of image noise. After that, potential edges
are reduced to 1-pixel curves by eliminating non-maximum pixels of the gradient magnitude. Finally,
edge pixels are retained or deleted using hysteresis thresholding on the gradient magnitude. The
Gaussian width is the width of the Gaussian curve (the noisier the picture, the wider the width), as
well as the low and high thresholds for hysteresis thresholding, are all customizable in the Canny.

Fig 10.
Canny filter Image

22
E. Sobel Edge Detection Filter

The Sobel operator, also known as the Sobel–Feldman operator or Sobel filter, is a tool used in image
processing and computer vision, particularly for edge identification. It is named after Stanford
Artificial Intelligence Laboratory colleagues Irwin Sobel and Gary Feldman (SAIL). This method
carry-out a 2-D spatial gradient measurement on a picture that helps in emphasizing high-frequency
regions that correspond to edges. It is used to calculate the approximate absolute gradient magnitude
at each point in a grayscale image input. For Edge Detection, the Sobel Filter is utilized. It calculates
the picture intensity gradient at each pixel inside the image. It determines the direction and pace of
change in the most significant rise from light to dark. The Prewitt operator and the Sobel operator are
extremely similar. It's also a derivate mask that's utilized to identify edges. The main distinction is
that with the Sobel operator, the mask coefficients are not set and may be altered according to our
needs unless they break some derivative mask condition.
When we apply this mask to an image, the vertical edges become more visible. It simply calculates
the difference of pixel intensities in an edge area, similar to a first order derivate. Because the center
column is 0, it does not contain the image's original values, instead calculating the difference between
the right and left pixel values around that edge. In addition, the first and third column's center values
are 2 and -2, respectively. This gives the pixel values at the edge of the display more weight. The
edge intensity is increased, and the picture is boosted in comparison to the original.

Fig 11.
Sobel Filter Image

23
3. Pre-Processing

In the pre-processing part we have used median, contrast and negate filter which will process our
ultrasound images through these filters.
The following images will show us the implementation of the project-

Fig 12.

Negate Filter in Pre-processing

24
 Fig 13.

Contrast Filter in Pre-processing

25
 Fig 14.

MATLAB code design in Pre-processing

26
Code used in MATLAB to apply filters in the Pre-Processing step

b=get(handles.popupmenu1,'value');
a=handles.a;
if(b==1)
L = 2 ^ 8;
neg = (L - 1) - a;
axes(handles.axes1);
imshow(neg);
e=entropy(neg);
set(handles.edit5,'string',e);
elseif(b==2)
eps = 0.1;
W = 1 ./ (1+ (100 ./(double(a) + eps)) .^ 20);
axes(handles.axes1);
imshow(W);
e=entropy(W);
set(handles.edit5,'string',e);
elseif(b==3)
c =0.1;
V = c * log (1 + double(a));
axes(handles.axes1);
imshow(V);
e=entropy(V);
set(handles.edit5,'string',e);
elseif(b==4)
Y = imadjust (a, [0 1], [1 0], .5);
axes(handles.axes1);
imshow(Y);
e=entropy(Y);
set(handles.edit5,'string',e);
end

27
ALGORITHM

• Load the input images from the available database.

• Applying pre-processing filters on images such as negate, contrast, median filter that are
widely used in the field of image processing.
• We have then chosen/customized our filter values and their sensitivity according to our
requirement.
• We have then loaded and processed our image on console and obtain the values of each image
that we have generated followed by further processing of image in the post-processing step.

28
4. Post Processing
In the Post-Processing part we have used Edge detection filters like Sobel, Canny and Prewitt filter
which will process our ultrasound images from the dataset through these edge detection filters.

The following images will show us the implementation of the project-

Fig 15.

Sobel Filter in Post-processing

29
 Fig 16.

Canny Filter in Post-processing

30
 Fig 17.

Prewitt Filter in Post-processing

31
 Fig 18.

MATLAB code design in Post-Processing

32
Code used in MATLAB to apply filters in the Post-Processing step

b=get(handles.popupmenu2,'value');
a=handles.a;
resized_image=imresize(a,[256 256]);
I=0.21*resized_image(:,:,1) + 0.72*resized_image(:,:,2) + 0.07*resized_image(:,:,3);
if(b==1)
BW1 = edge(I,'sobel');
axes(handles.axes1);
imshow(BW1);
e=entropy(BW1);
set(handles.edit5,'string',e);
elseif(b==2)
BW2 = edge(I,'Prewitt');
axes(handles.axes1);
imshow(BW2);
e=entropy(BW2);
set(handles.edit5,'string',e);
elseif(b==3)
BW3 = edge(I,'Canny');
axes(handles.axes1);
imshow(BW3);
e=entropy(BW3);
set(handles.edit5,'string',e);
end

33
ALGORITHM

• Load the input images from the pre-processed image database.

• Applying post-processing filters on images such as sobel, canny and other edge detection
customised filters that are widely used in the field of image processing.
• We have then chosen/customized our filter values and their sensitivity according to our
requirement.
• We have then loaded and processed our image on console and obtain the values of each image
that we have generated followed by further processing of image and then pixel values has been
recorded for further analysis.

34
5. Prediction

Based on the pre-processing and post processing of data, we have obtained the prediction on the
presence of kidney stones using ultrasound images and the prediction button will analyse our applied
filters and their values passed on every pixel of the selected image from the database and if any changes
in the value occurs, the system will predict the possibility of unidentified objects in the kidney using
ultrasound images
The following images will show us the implementation of the project-

Fig 19.

After Pre-processing and Post-Processing of the image

35
 Fig 20.

In this, No Stone is detected After Pre-processing and Post-Processing of the image

36
 Fig 21.

Similarly, Stone is detected in the different image After Pre-processing and Post-Processing of the
input image from database

37
 Fig. 22

MATLAB code design for Prediction

38
Code used in MATLAB for making Prediction after processing

a=handles.a;
b=rgb2gray(a);
c=b>20;
d=imfill(c,'holes');
e=bwareaopen(d,1000);
PreprocessedImage=uint8(double(a).*repmat(e,[1 1 3]));
PreprocessedImage=imadjust(PreprocessedImage,[0.3 0.7],[])+50;
uo=rgb2gray(PreprocessedImage);
mo=medfilt2(uo,[5 5]);
po=mo>250;
[r c m]=size(po);
x1=r/2;
y1=c/3;
row=[x1 x1+200 x1+200 x1];
col=[y1 y1 y1+40 y1+40];
BW=roipoly(po,row,col);
k=po.*double(BW);
axes(handles.axes1);
imshow(k);
M=bwareaopen(k,4);
[ya number]=bwlabel(M);
if(number>=1)
h= msgbox('Stone is Detected','Sucess');
object_handles = findall(h);
set(h, 'position', [200 60 200 60]);
set( object_handles(6),'FontSize', 20, 'HorizontalAlignment', 'left','VerticalAlignment', 'bottom');
else
h=msgbox('No Stone is detected','Failed','error');
object_handles = findall(h);
set(h, 'position', [200 60 200 60]);
set( object_handles(6),'FontSize', 20, 'HorizontalAlignment', 'left','VerticalAlignment', 'bottom');
end

39
 6. Accuracy
Based on the Prediction we need to find the accuracy of the system so that we could compare our results
with the existing approach and find the efficiency of our project which can help us to implement this
project in the real world.
The following are the images we obtained for accuracy in our project –

Fig 23.

Accuracy of the particular image after prediction is 83.9021

40
Code used in MATLAB for Calculating and reflecting Accuracy on the Predicted Image

a=handles.a;
b=rgb2gray(a);
img=b;
img=double(img(:));
ima=max(img(:));
imi=min(img(:));
mse=std(img(:));
snr=120*log10((ima-imi)./mse);
msgbox(sprintf('Accuracy:%0.4f',snr));

41
 7. Sensitivity
Based on the Prediction, we need to find the sensitivity of the system so that we could compare our
results with the existing approach and find the efficiency of our project which can help us to implement
this project in the real world.

The following are the images we obtained for accuracy in our project –

Fig 24.

Sensitivity of the particular image after prediction is 80.4062

42
Code used in MATLAB for Calculating and reflecting Accuracy on the Predicted Image

a=handles.a;
b=rgb2gray(a);
img=b;
img=double(img(:));
ima=max(img(:));
imi=min(img(:));
mse=std(img(:));
snr=115*log10((ima-imi)./mse);
msgbox(sprintf('Senstivity:%0.4f',snr));

43
 CHAPTER 04

PERFORMANCE ANALYSIS

4.1 Discussion on the Results Achieved

Based on the Prediction we made in the project we also obtained the accuracy and sensitivity values
for each image from the database and we have recorded randomly 10 iterations from the database to
generate mathematical analysis of our project .

The following table is created using MS-Excel when we recorded data for 10 iterations that shows
kidney stones in ultrasound images-

S.No Pre-Processing Post-processing

Parameter Accuracy Senstivity Accuracy Senstivity

1 82.5 80.3 80.2 80

2 89.4 84.3 86.2 82.3

3 87 84.1 84 83.7

4 87.2 86.4 82.4 84.6

5 91.2 87.2 84.1 85

6 87.2 88.5 84.2 82.1

7 89.6 84 86.5 85.2

8 85.8 81.5 81.8 79.3

9 86.3 84.6 82.1 83.8

10 84.5 82.1 80.5 80.1

Table 2.

44
Based on the data recorded from images on MS-Excel, We further calculated the accuracy and
sensitivity of our project and also a comparison between the Pre-Processing Filters and Post-
Processing Filters .

The following output is received for ultrasound where kidney stone is detected –

No.of Std. Mean

GROUP Samples Mean Std. Deviation Error

Pre-

Processing 228 87.07 2.552145067 0.807059133

Post-

Accuracy Processing 228 83.2 2.171532997 0.686699028

Pre-

Processing 228 84.3 2.568181371 0.812130258

Post-

Sensititvity Processing 228 82.61 2.196183558 0.69449422

Table 3.

45
The following graphs compares the accuracy and sensitivity of the Pre-Processing and Post-
Processing based on 10 iterations recorded of ultrasound images where kidney stones are detected

Accuracy Comparison
95

90

85

80

75

70
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Pre-Processing Pre-Processing

Sensitivity Comparison
90

88

86

84

82

80

78

76

74
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Post-Processing Post-Processing

Fig 25.
Graphs representing Accuracy and sensitivity

46
The following table is created using MS-Excel when we recorded data for 10 iterations that does not
detected kidney stones in ultrasound images-

S.No Pre-Processing Post-processing

Parameter Accuracy Senstivity Accuracy Senstivity

1 84.5 84.7 80.2 85.2

2 88.4 84.3 85 88.6

3 82 84.1 84.2 84.3

4 86.4 86.4 84.2 84.6

5 81.1 83.1 85.3 85

6 87.2 84.6 88 82.1

7 89.6 86.2 84.9 85.2

8 85.7 88.9 86.5 79.3

9 88.1 84.6 87.8 76.2

10 89.1 83 84.7 88.1

Table 4.

47
Based on the data recorded from images on MS-Excel, We further calculated the accuracy and
sensitivity of our project and also a comparison between the Pre-Processing Filters and Post-
Processing Filters .

The following output is received for ultrasound where kidney stone is not detected –

No. of Std. Mean

GROUP Samples Mean Std. Deviation Error

Pre-
86.21 2.910689
Processing 228 0.807059133

Post-
85.08 2.199394
Accuracy Processing 228 0.686699028

Pre-
84.99 1.761596
Processing 228 0.812130258

Post-
83.86 3.780711
Sensititvity Processing 228 0.69449422

Table 5.

48
The following graphs compares the accuracy and sensitivity of the Pre-Processing and Post-
Processing based on 10 iterations recorded of ultrasound images where kidney stones are detected

Accuracy Comparison
92
90
88
86
84
82
80
78
76
74
1 2 3 4 5 6 7 8 9 10

Pre-Processing Post-processing

Sensitivity Comparison
90

85

80

75

70

65
1 2 3 4 5 6 7 8 9 10

Pre-Processing Post-Processing

Fig 26.
Graphs representing Accuracy and Sensitivity

49
4.2 Project Outcome:

• Project Accuracy: 87%.

• Project Sensitivity: 84%
• Project Report: Completed
• Research Paper: Submitted
• Website: N/A
• Patent: N/A

50
 CHAPTER 5

5.1 Conclusion:
We, in this Major Project, Developed a Kidney stone Detection system using Image Processing
Techniques. We Initially started with a database of 114 ultrasound images and further interest to the
project we obtain another 114 images from a hospital’s website. we Initially wanted to develop a model
with the highest accuracy and compete with the existing AI and Deep learning models present in the
industry with similar kind of project. We used several Image Processing Techniques including (Log
filter, Negate filter, Contrast Filter, Edge detection techniques such as Prewitt filter, Sobel Filter and
Canny filter. From our calculations and recorded data, we were able to:
• Achieved 87% Accuracy using Basic Digital Image Processing Techniques.
• Create a user friendly model which can be implemented by anyone
• Able to utilize ultrasound images and make prediction based on that which is itself beneficial
in terms of cost as well as easy to get to any patient or doctor.

51
5.2 Applications of the Major Project

• This project can be widely used in the field of healthcare.

• This project can also be widely used by any person who is having any doubt regarding kidney
problem or is suffering from some kidney disease.
• This project can be used by researchers as well to detect new object inside any hollow or blurred
image using basic image processing techniques.
• This project can be used by doctors / medical practitioner for operating kidney stones related
operations.
• This project can be used by researchers for studying about kidney and kidney stones related
diseases for improving their research in the field of medicine.

52
5.3 Limitations of the Major Project:

Note that although in our Project we have got a 87 % prediction accuracy, we are still researching on deep
learning and neural network methods corresponding to this project as they can work better and more
efficiently if the data set in hand was more complex and bigger. However, the study's fundamental flaw is
the healthcare system's complexities. These limitations will be overcome over time, but in the meanwhile,
we will be able to apply more complex combinations of ways to achieve more significant results.

It takes more time to take load more complex dataset i.e images of high pixel with different properties to
run on the system and extract features from the images.

53
5.4 Contributions:

• Nishkarsh Sharma:
Research and Development for the Database, GUI Development, Image Pre-
Preprocessing, Image Post-Processing Research and development of approach taken for
Prediction, Calculating Accuracy, Tracking Sensitivity, Data Analysis, Comparison and
visualization of data.

54
5.5 Future Work

● Note that although in our Project we have got a 87 % prediction accuracy, we are still
researching on deep learning and neural network methods corresponding to this project as they can
work better and more efficiently if the data set in hand was more complex and bigger.

● First, we are thinking of increasing the size of the current data set by a huge amount by gathering
dataoff the internet or by putting some random manual data into it, then we will try to implement a
Neural Network for Kidney stone Detection using deep learning and Artificial Intelligence Model, as
of right now we are in the process of learning and researching about neural network and its different
implementation on this project.

55
References

56
57
58