International Journal of Multidisciplinary Research and Development 
 

International Journal of Multidisciplinary Research and Development

Online ISSN: 2349-4182 Print ISSN: 2349-5979, Impact Factor: RJIF 5.72
www.allsubjectjournal.com
Volume 3; Issue 5; May 2016; Page No. 81-83

Features extraction and classification for detection of kidney stone region in ultrasound images
1
Monika Pathak, 2 Harsh Sadawarti, 3 Sukhdev Singh
1
Research Scholar, I. K. Gujral Punjab Technical University, Jalandhar, Punjab, India.
2
Director, Department of Computer Science, RIMT, Mandi Gobindgarh, Punjab, India
3
Assistant Professor, Computer Science Department Multani Mal Modi College, Patiala, Punjab

Abstract
Region of interest detection in ultrasound image is a challenging task due to heterogeneous texture and presence of speckle noise.
The ultrasound scanning is most frequently used tool to examine the patient for abnormities, especially presence of stone, in the
kidney. Automatic object detection in ultrasound images is burning research areas and the present research work is in the same
direction. We have developed an application, which helps the medical practitioner to identify the stone region in the ultrasound
image. It is a semiautomatic system in which practitioner need to select the region, which is analyzed, by the proposed system for
presence of stone. The feature extraction is applied on cropped regions, which may contain stone. The various features such as
Contrast, Angular second moment, Entropy and Correlation are used. The KNN classifier is used to classification based on training
image dataset. The overall accuracy of classification system is around 91%. The confusion matrix is also prepared to analyze the
complexity and accuracy of the proposed system.

Keywords: feature extraction, KNN classifier, Kidney stone, confusion matrix, statistical features
1. Introduction
Kidney ultrasound imaging is an economical, non-invasive, abnormalities in the shape, size and location of the kidneys in
real time diagnosis system, which is used to measure the body [1].

Fig 1: Ultrasound Images show Kidney stone and shadow

As shown in the figure1, the ultrasound B-mode image shows and classification.
oval shape kidney with well-defined boundary. On the lower
right side of the kidney, there is round object, which is expected 2. Features extraction
to be a stone. The visual characteristic of a stone in ultrasound The objective of the features extraction is to capture important
image is the presence of shadow. The black shade immediately characteristics of region under investigation in the kidney
after the round object confirms the presence of stone. However, image [2]. The features of the region should be able to identify
in case of machine learning, we look for features like shade, region uniquely. The present research has used statistical
contrast; correlation etc. to identify the region contains the features to identify the region of interest. The concept of gray
stone. level co-occurrence matrix (GLCM) used to extract the
The present research work is focused on object (stone) following features [3].
identification in ultrasound images based on feature extraction

Table1: Description of statistical features

S. No. Feature Description
It defines the difference between the lightest and darkest areas on an image
1. Contrast ,

Where i and j are the pixel values.

81 
 
 
International Journal of Multidisciplinary Research and Development 
 

It is the state or quality of being homogeneous. It is calculated as sum of square of angular entries
in GLCM moments. The higher value of ASM indicates textural uniformity.
Angular second
2. moment (ASM,
Energy) ,

Where, Ng is gray tone image in GLCM form.

Entropy measures the randomness of the image texture (intensity distribution).

, ,
3.
Entropy The homogeneous image shows lower entropy value, whereas, heterogeneous region results in a
higher entropy value
Correlation is a measure of the strongest of the relationship between two variables.
4. Correlation ,

The following features are extracted from the GLCM of the ROI
kidney images using MATLAB [4, 5, 6, 7]. Energy, Entropy, Contrast,
Homogeneity, Maximum probability and correlation.

Table 2: Sample of Features Extracted from Ultrasound Kidney

stone Images The KNN machine-learning program takes decision based on
Image 1 Image 2 previously stoned decisions. The basic principle of KNN is
Feature based on the rule to majority vote to its neighbors [9]. The
Min Value Max Value Min Value Max Value
Contrast 0.8842 1.3792 0.8991 1.2290 present research used it to classify the kidney stone images by
ASM, (Energy) 0.1902 2.0127 0.2811 2.7654 comparing the features extracted from training dataset.
Entropy 2.3754 2.5419 1.9251 2.4019
Correlation 0.5452 0.7172 0.3912 0.4138 Training Phase
The kidney Ultrasound images are collected and stone regions
3. Classification cropped to create training dataset. The training dataset contains
The k-Nearest Neighbors algorithm (or k-NN) is a nearly 250-cropped images, which contain stone. The feature
nonparametric algorithm, which is used to classify different extraction is applied on the training image dataset and features
cases based on some similarity measures such as Euclidean are stored as knowledge base. Table3 shows few images from
distance. The Euclidean distance is measured using following training dataset.
equation [8].
Table 3: Sample images of training image dataset

82 
 
 
International Journal of Multidisciplinary Research and Development 
 

The features extracted in the feature extraction phase are stored 7. References
for classification analysis. The KNN classifier analyses the 1. Loganayagi T, Kashwan KR. A Robust Edge Preserving
new feature set with the previously stored features set. If the Bilateral Filter for Ultrasound Kidney Image. Indian
features are close to the previous set of features then the image Journal of Science and Technology. 2015; 8(23).
is well classified otherwise the image is rejected for stone 2. Bojunga J, Dauth N, Berner C, Meyer G, Holzer K, Voelkl
images. The training image set show significant results and L. et al. Friedrich-Rust, Acoustic Radiation Force Impulse
well classifies the stone images. Imaging for Differentiation of Thyroid Nodules, Plos One,
2012; 7(8).
4. Result and Analysis 3. Tamilselvi, Thangaraj. Computer Aided Diagnosis
The statistical analysis of the present algorithm carried out on System for Stone Detection and Early Detection of Kidney
kidney ultrasound images. The following statistical parameters Stones, Journal of Computer Science, 2011; 7(2):250-254.
are used for analysis: 4. Jenho Tsao, Li-Hsin Chang, Chia-Hung Lin. Ultrasonic
 Confusion Matrix: It contains information about actual Renal-Stone Detection and Identification for
and predicted classifications done by a KNN classifier [10]. Extracorporeal Lithotripsy, in the proceedings of the IEEE
The performance of the proposed region of interest Engineering in Medicine and Biology 27th Annual
detection system evaluated using the analysis of the Conference Shanghai, China, 2005, 1-4.
confusion matrix. Table4 shows the confusion matrix for a 5. Chalekar P, Shroff S, Pise S, Panicker S. Use of k-Nearest
two-class classifier. Neighbor in Thyroid disease Classification. Technical
Research Organization India. 2014; 1(2):36-41.
Table 4: Confusion Matrix 6. Grigorescu SN, Petkov N, Kruizinga P. Comparison of
Predicted Texture Features Based on Gabor Filters, IEEE
Negative Positive Transactions on Image processing, 2002; 11(10).
Negative TN FN 7. Pal M, Mather PM. A Comparison of Decision Tree and
Actual Backpropagation Neural Network Classifiers for Land
Positive FP TP
Use Classification IEEE International Symposium on
 Accuracy: The accuracy of classification process is based Geoscience and Remote Sensing, 2002, 503-506.
on correct and incorrect predictions. Following formula 8. Selvathi D, Sharnitha VS. Thyroid classification and
used to calculate the accuracy of the classification process segmentation in ultrasound imaging using machine
[11]
. learning algorithm international conference on signal
processing, communication, computing and network
techniques (ICSCCN), 2011.
9. Suganya R, Rajaram S. Content Based Image Retrieval of
Where, Ultrasound Liver Diseases Based on Hybrid Approach.
TP (True Positive) – number of correct positive prediction, American Journal of Applied Sciences. 2012; 9(6):938-
TN (True Negative) – number of correct negative Prediction, 945, ISSN 1546-9239.
FN (False Negative) – Number of incorrect negative 10. Sridhar S. Segmentation of Ureteric and Bladder Calculi
predictions. in Ultrasound Images. Journal of Computer Science. 2012;
FP (False Positive) – number of incorrect positive prediction. 8(5):716-720.
11. Maanasa NAS, Gowri V. Segmentation of mammogram
5. Conclusion using Tumour cut algorithm International Journal of
The present research work is an attempt to automate the Engineering and Innovative Technology (IJEIT), 2013;
Ultrasound analysis process for detection of stone. The system 2(10).
aimed to identify the ultrasound image region, which may
contain the stone. The stone region identification based on
analysis of stone region features. The system consists of feature
extraction and classification where the feature extraction
carried out on the ultrasound image data and testing image
dataset. The KNN classifier used to analyze the features for
detection of stone in the images. The overall accuracy of the
system is 91% which is satisfactory at the time. The accuracy
may be enhance by considering more suitable features which
can define stone region and multiple classifiers can also be
added into the system.

6. Acknowledgement
Authors acknowledge the opportunity and support provided by
I. K. Gujral Punjab Technical University, Jalandhar to conduct
the present research work.

83