Kidney Stones Detection Using Image Processing and BPNN

Shuvashis Shome Shuvo1, Arzun Kumar2, Md.Mayn Uddin3

1
Student,Department of Electrical & Electronic Engineering,Jatiya Kabi Kazi Nazrul Islam
University,Trishal,Mymensingh
2
Student,Department of Electrical & Electronic Engineering ,Jatiya Kabi Kazi Nazrul Islam
Univertsity,Trishal,Mymensingh
3
Assistant Professor,Department of Electrical & Electronic Engineering,Trishal,Mymensingh

[email protected]

[email protected]

[email protected]

Abstract
Kidney stone is one kind of urological disease which is found in the upper urinary tract.. This problem can
lead to many complications .It has also very high recurrence rate.If not detected at early stage, it can result
in renal failure or even permanent organ failure.So early detection of kidney stone is very vital.Identification
of true size and exact location of kidney stone is vital to successfully perform the surgery to remove stones.
Ultrasound imaging is widely known technique for kidney stone detection.At early stage kidney stone can
be got unnoticed using Ultrasonic or CT scan imaging as these images have very high level of unwanted
noise and artifacts.This paper presents an image processing technique for kidney stone detection.Acquired
CT and Ultrasonic images of kidney are first preprocessed.Then this preprocessed images are segmented
using k-means algorithm and watershed functioin.Morphollogical operation is performed to specify the
region of interest.Different features are extracted using glcm(gray level co-occurnce matrix)
algorithm.Back Propagation Neural Network(BPNN) is used to classify the images and increase the
accuracy of the detection.This detection technique is more accurate compared to visual inspection of the
physicians.Moreover it gives very high true positive rate than ultrasonic imaging .It also successfully
overcomes the challenges of handling large number of images for detection of kidney stones especially at
early stage.

I.Introduction:
Presently,number of people suffering kidney stones are increasing alarmingly.Kidney stones are identified
in the upper unirary tract.At the early stage,most of the people can not notice kidney stones in the kidney
which can cause damages to the organs of the body.To meet this this problem The automated detection of
kidney stone is important [1],[2]. Ultrasound imaging which is non-invasive, less costly process is widely
known technique for kidney stone detection [3]. Un noticed stone in kidney can lead to infection and renal
failure in the long run.To prevent renal failure,detection of kidney stone is vital as early as possible
[3].Recent research has demonstrated that stones in kidney is identified more easily using digital processing
technique than naked eye.It is highly tough for a physician to diagnosekidney stones from ultrasonic
imaging due to presence of noisy artifacts.The contrast resolution of ultrasonic images is very limited which
makes it hard to carry out accurate diagnostic performance. This paper presents a digital image processing
technique to enhance the quality of ultrasonic image by image filtering segmentation and morphological
operation to specify the region of interest according to [4], [5],[6],[7],[8].
A Backpropagation Neural Network (BPNN) is used to enhance the performance and accuracy level of the
detection as mention [11],[12]. GLCM is used to extract feature and analysis as demonstrated in [9], [15].

II.Literature Review:
Using Digital Image processing technique,acquired images are preprocessed and Region of
Interest(ROI) is identified as mentioned in [8]. A.McAndrew demonstrated in [10] a digital image
processing technique using MATLAB software for medicall imaging.W.M.Hafizah mentioned in [16] a
feature extraction method for ultrasound images of kidney using glcm(gray level co-occurrence matrix).N.
Koizumi demon strated in [17] a non- invasive ultrasaound diagnostic system for tracking of kidney
stone.Attia MW mentioned in [13] a classification model for ultrasound kidney images using neural
network and PCA.

III.Image processing Methods and Materials:

This section derives our proposed approach for the detection of kidney stones using image
processing techniques and back propagation neural network(bpnn).

Image Image pre- Image Morphologi

acquisition processing segmentation cal analysis

Image
Feature
classification
extraction
using BPNN

Fig:Image processing technique for our proposed approach

classification u sing
A.Image Acquision:
This is the fast task for image processing.Several ultrasonic images were taken to perform the process
from Kaggle.

B. Image pre-processing:
It is the simulation section of image processing.It includes RGB to grayscale conversion of ultrasonic
image.Then it is converted grayscale to binary using Matlab coding.During binarization binary image is
filled with holes.Thus image is done.It is followed by median filtering and noise removation.Thus quality
of image is improved significantly .

C. Image segmentation:
At this stage images are segmented using k-means algorithm. Region of interest(ROI) is selected while
removing noice and distortion.Kidney stone is located at the ROI.The figure shown below display the
segmented image containing ROI.K-means clustering algorithm is used to implement this process.
D.Morphological Analysis:
Morphological operation of image processing refers the image enhancement.Here dilation and erosion
performed.Intensity of pixel lower than 250 is eliminated for morphing the images. The entire image
processing is performed using Matlab(R2018a) software.

E.Image Classification Using Backpropagation Neural Network:

In machine learning,for classification backpropagation neural network(bpnn) is extensively used algorithm
.for classifying our model,feed forward back propagation neural network is used. In order to fit a
network,gradient of loss function is computed with respect to weighted sum for a single input network.Thus
efficiency is increased and the loss is minimized [14].Accuracy and performance of a network is
significantly increased using BPNN.
F. Feature Extraction:

To attain high accuracy level,feature extraction is done .Gray-level co-occurance matrix is used to extract
different feature based features.
IV. GLCM:

GLCM or Gray level co-occurance matrix is widely used for feature extraction and statistical
distribution of features several texture based like energy,correlation,entropy,homogenerity etc. are
extracted using GLCM.
Energy: Energy also known as ‘angular’ second moment.It is proportional to image pixel frequency.
𝑁−1
Energy= ∑ (𝑃𝑦 )2
𝑖,𝑗=0

Entropy:Entropy refers image randomness a lower entropy value indicates a homogenous image.

𝑁−1
Entropy=
∑ − ln(𝑃𝑖𝑗 ) 𝑃𝑖𝑗
𝑖,𝑗=0

Contrast: It determines linking intensity of a pixel with respect to its neighbour.

𝑁−1
Contrast= ∑𝑖,𝑗=0(𝑖 − 𝑗)2

𝑁−1
(𝑖−µ)(𝑗−µ)
Correlation: Correlation of a pixel with it’s neighbour correlation= ∑ 𝜎2
𝑖,𝑗=0
Homogenity:Similarity of GLCM matrix of a homogenous image and pixels is denoted by homogeneity.
𝑁−1 𝑝𝑖𝑗
Homogeneity= ∑ 2
𝑖,𝑗=0 1+((𝑖−𝑗)

V.Result and discussion:

Total 34 Ultrasound and CT scan images are taken from Kaggle datasets for implementation of this
project.From the dataset,15 images are identified as normal and 19 images are identified as stone image.

Entire experimental work is carried out using matlab software.Device configuration includes Intel i5
processor,windows 10(64bit),RAM 8GB.
Figure 2:Pre-processing results for image 1
Results:Stone is detected
Figure 3: Pre-processing results for image 2(Stone is not detected)

Table 1:Feature extraction:Table 2(GLCM features values)

Feature Image1 Image2
Contrast 0.1021 0.0590
Correlation 0.9519 0.9843

Performance 4.7842e-05 0.004585

Energy 0.5750 0.3955
Entropy 0.5859 0.8571
Homogenity 0.9777 0.9733
5.Image Classification using Back Propagation(BPNN) Neural Network:
For classification purpose, Back Propagation Neural Network(BPNN) is performed. Datas are
collected from Kaggle dataset

Performance Validation Graph:

Figure 4: Performance Validation graph:
Performance validation graph is shown in the figure 4. X-axis represents number of epochs
needed to perfom this model.Y-axis represents cross-entropy value.Performance graph of train
and validation data is near to the optimal performance range.However, performance graph of
test data is slightly under the optimal range. So it is proved that performance of our proposed
Neural Network is very accurate.
Confusion Matrix:

Figure of Confusion Matrix

From the confusion matrix ,it is clearly seen that images are successfully classified.
Total number of image:34
Classification accuracy:100%
Misclassification rate:0%
 VII: Conclusion:
This project has successfully detected kidney stones in the ultrasound images.Backpropagation Neural
Network (BPNN) and GLCM feature extraction has enhanced performance,accuracy and efficiency.It has
made easier to diagnose kidney stones as compared to visual impaction. Entire simulation is done using
MATLAB R2018 SOFTWARE. At the early stage many kidney stones get unnoticed to the visual
inspection.Image processing along with classification can correctly detect early stage kidney stone . It has
also overcome some of the challenge including handling huge number of images.This method can be
further developed for various medical image classification purpose.A machine learning based automated
detection model can be developed to predict the size and number of stones in kidney.
Keywords: Kidney Stone , Image Processing Technique , BPNN, Feature Extraction

References:
[1] A. I. Nouri and M. A. Hassali, “Assessment of kidney stone disease prevalence in a teaching hospital,”
African Journal of Urology, vol. 24, no 3, pp 180-185, 2018.

[2] W. Brisbane, M. R. Bailey, and M. D. Sorensen, “An overview of kidney stone imaging techniques,”
Nature Reviews Urology, vol. 13, no 11, pp 654-662, 2016.

[3] W. M. Hafizah, and E. Supriyanto, “Automatic generation of region of interest for kidney ultrasound
images using texture analysis,” International Journal of Biology and Biomedical Engineering, vol. 6, no 1,
pp 26-34, 2012.

[4] T. Vineela, R. V. G. L. Akhila, T. Anusha, Y. Nandini, and S. Bindu, “Kidney Stone Analysis Using
Digital Image Processing,” International Journal of Advanced Research in Electrical, Electronics and
Instrumentation Engineering, no 3, pp 001-004, 2020.

[5] Bamber JC, Daft C. ‘‘Adaptive filtering for reduction of speckle in ultrasonic pulse-echo images.
Ultrasonics’’. 1986 Jan 1; 24(1):41-4.

[6] Loupas T, McDicken WN, Allan PL. ‘‘ An adaptive weighted median filter for speckle suppression in
medical ultrasonic images. IEEE transactions on Circuits and Systems’’. 1989 Jan; 36(1):129-35.

[7] Sudha S, Suresh GR, Sukanesh R. ‘‘ Speckle noise reduction in ultrasound images by wavelet
thresholding based on weighted variance. International Journal of Computer Theory and Engineering’’.
2009 Apr 1; 1(1):7.

[8] Gai S, Zhang B, Yang C, Yu L, ‘‘ Speckle noise reduction in medical ultrasound image using
monogenic wavelet and Laplace mixture distribution. Digital Signal Processing’’. 2018 Jan 1; 72:192-207.

[9] Hafizah WM, Supriyanto E, Yunus J, ‘‘ Feature extraction of kidney ultrasound images based on
intensity histogram and gray level cooccurrence matrix’’. In 2012 Sixth Asia Modelling Symposium 2012
May 29 (pp. 115-120). IEEE.

[10] A. McAndrew, “An Introduction to Digital Image Processing with MATLAB Notes for SCM2511
Image Processing’’. 1, Semester 1, 2004.
[11] Subramanya MB, Kumar V, Mukherjee S, Saini M, ‘‘ SVM-based CAC system for B-mode kidney
ultrasound images. Journal of digital imaging’’. 2015 Aug 1; 28(4):448-58.

[12] Vinoth R, Anusha C, SaiPriya D, Sridhar BS, ManiKantaPavan C, Aditya G. ‘‘ A Novel Approach
on Ultrasound Kidney Images Classification by using Micro Deep Neural Network Algorithm. International
Journal of pure and applied Mathematics’’. 2018; 119 (12):15047-15058.

[13] Attia MW, Abou-Chadi FE, Moustafa HE, Mekky N, ‘‘Classification of ultrasound kidney images
using PCA and neural networks. International Journal of Advanced Computer Science and Applications’’.
2015 Apr 1; 6(4):53-7.

[14] Raja KB, Madheswaran M, Thyagarajah K. ‘‘ A hybrid fuzzy-neural system for computer-aided
diagnosis of ultrasound kidney images using prominent features. Journal of medical systems’’. 2008 Feb 1;
32(1):65-83.Engineering, Science and Management (ICAESM '12), pp. 601–605, March2012.
[15]. W. M. Hafizah, “Feature extraction of kidney ultrasound images based on intensity histogram and
gray level cooccurrence matrix,” in Proceedings of the 6th AsiaModeling Symposium (AMS ’12), pp.
115–120, IEEE, May 201

[16] W. M. Hafizah, “Feature extraction of kidney ultrasound images based on intensity histogram and
gray level co-occurrence matrix,” in Proceedings of the 6th Asia Modeling Symposium (AMS ’12), pp.
115–120, IEEE, May 2012 62

[17] N. Koizumi, J. Seo, D. Lee et al., “Robust kidney stone tracking for a noninvasive ultrasound
theragnostic system—servoing performance and safety enhancement,” in Proceedings of the IEEE
International Conference on Robotics and Automation (ICRA '11), pp. 2443–2450, Shanghai, China,
May2011.

Authors’s Bio:
Author1:Shuvashis Shome Shuvo

Student,Department of Electrical & Electronic Engineering

Institution:Jatiya Kabi Kazi Nazrul Islam University,Trishal,Mymensingh

Mobile:01783540200

E-mail:[email protected]

Author2:Arzun kumar
Student,Department of Electrical & Electronic Engineering

Institution:Jatiya Kabi Kazi Nazrul Islam University,Trishal,Mymensingh

Mobile:01755147818

E-mail: [email protected]
Author3: Md.Mayn Uddin
Assistant Professor,Department of Electrical & Electronic Engineering,Trishal,Mymensingh

Institution:Jatiya Kabi Kazi Nazrul Islam University,Trishal,Mymensingh

Mobile:01812911488

E-mail:[email protected]