Optimal Feature Selection and Classification of Carotid Artery Images

DOI : 10.17577/IJERTV8IS100062

Download Full-Text PDF Cite this Publication

Text Only Version

Optimal Feature Selection and Classification of Carotid Artery Images

Asst. Prof. Ms. B. Farhana Ansoor

Asst. Prof. Mr. I. Mohammed Farook

Department of Biomedical Engineering

Department of Electronics and Communication Engineering

Aarupadai Veedu Institute of Technology

Aalim Muhammed Salegh College of Engineering

Chennai, India

Chennai, India

Abstract – Cholesterol plaque may accumulate gradually in the carotid artery wall. This growing plaque may eventually narrow the carotid artery and reduces blood flow which leads to stroke, heart attack, and peripheral vascular disease. In this paper, we proposed a technique to easily detect a disease which is present in the carotid artery. Features like Spatial Wavelets, Moment of Gray-Level Histogram(MLGH) and Gray Level co- occurrence Matrix(GLCM) are extracted from ultrasound images of the carotid artery, and the extracted feature set is optimized using the genetic algorithm. To detect the presence of plaque in the carotid artery, the ultrasound image is segmented using improved Spatial FUZZY C means algorithm. The IMT (Intima Media Thickness) value is measured from the Segmented image. With the help of optimized feature set and IMT value, the images are classified into normal or abnormal using MLPBNN(Multi-Layer Back-Propagation Neural Networks).



Feature Extraction

Feature Extraction

Index Terms Plaque; MGLH; GLCM; FUZZY C means; IMT; MLBPNN.

  1. OVERVIEW OF THE PROPOSED SCHEME Enhancement of an image using histogram equalization

    and removal of noise using median filter is done in our proposed scheme. Features like MGLH, DWT, and GLCM are extracted from the de-noised image and the required features are optimized using genetic algorithm. Using FUZZY C means, the images is segmented. From the segmented images, the IMT values are measured and the image is classified using MLBPNN.


    Input image


    An ultrasound imaging technique is a non- invasive test used to visualize the organs with the help of sound waves. It is commonly used test and the computer constructs the images. The carotid arteries are the major blood vessels and it has two types namely internal and external carotid arteries. Internal carotid artery supplies oxygenated blood to the brain and external carotid artery supplies oxygenated blood to the face and neck. It is made up of 3 layers, smooth innermost layer called intima, muscular middle layer called media, and the outermost layer called adventitia. We can also detect the presence of plaque by measuring the thickness of intima-media called IMT measurement. Growth of plaque in the arteries may eventually narrows it, thus it reduces the blood flow. Heart attack if the thickness of plaque increases in the coronary artery, similarly stroke occurs if the thickness of plaque increases in the carotid artery.

    A. Pre-Processing

    A. Pre-Processing



    In pre-processing, the contrast of image is increased and the noises are removed from the image using filters. Extracting features like MGLH, DWT, and GLCM from the pre-processed image. Among the extracted features, the required features set is measured from the segmented image and the MLBPNN classifies the images into normal or abnormal.

    Feature Optimization

    Feature Optimization






    To improve the clarity and contrast of the image, removal of noise is required. Histogram Equalization is applied to enhance the intensity of an image. Generally an image will have an invisible unknown noise hence to remove it, addition of known noise must required. Thus, a known salt and pepper noise is added.

    Several filters are commonly used but we have used median filter [2] to de-noise an image because it is a non- linear digital filtering technique and it prevents the edges of an image whereas other filters smoothens the edges.

    1. Feature Extraction

      Feature Extraction is the technique used for obtaining necessary features of an image like Moments of Gray level Histogram (MGLH), 2DCWT, and Gray Level Co- occurrence Matrix (GLCM).

      1. Moment of Gray-Level Histogram

        A most frequently used technique for analyzing the texture of an image based on the properties of intensity histogram [3]. The moment of nth order about the mean is determined by the following expression,


        FM1 = (1)

        Standard Deviation

        FM2 = (2)


        FM3 = 1- (3)

        Third Moment

        FM4 = . (4)


        FM5 = (5)


        FM6 = (6)

        FM7 = (7)

        FM8 = (8)

        FM9 = (9)

      2. 2D-CWT

        Continuous wavelet transform is suitable for non- stationary signals. Robust processing is provided by this technique in pattern analysis [1]. Cover map concept has been put forward in order to speed up the

        analysis of 2D-CWT and to approach the computational time problem.

        The wavelet function is expressed as,


      3. Gray Level Co-occurrence Matrix (GLCM)

        The pixel spatial relationship is considered here, thus it is also termed as gray level spatial dependence matrix [5]. By default, the spatial relationship is defined as the pixel of interest and the pixel to its immediate right, but one can specify other spatial relationship such as vertical diagonal and off-diagonal, etc. between two pixels. In GLCM feature extraction, 5×5 window size is used. The information from the extracted feature is not reliable, when the size of window becomes too small. An erroneous textual information occurs, when the size of window becomes too large.

        Contrast = (11)

        Correlation = (12)

        Sum of Squares = (13)

        Inverse Difference (14)

        Sum Average (15)

        Sum (16)

        Sum (17)

        Entropy (18)

        Difference (19)

        Homogeneity (20)

        Dissimilarity= (21)

    2. Feature Selection

      Feature Optimization is done to improve the accuracy and computational time. In feature selection, necessary features are selected by using Genetic algorithm [4]. The irrelevant, noisy and redundant features are omitted to

      achieve good performance. There are many search methods available to select the features but we have used genetic algorithm because it reduces the classifier complexity and enhances its performance.

    3. Segmentation

      To classify the image pixel, five different types of clustering techniques are available but we have used FUZZY C means [8] because it has less influence to noise than many other clustering techniques. It is a very powerful method to segment a noisy image and it works effectively for both featured data

    4. Classification

    Images from different patients are collected and the features are distinguished by its values and it is classified as normal or abnormal. For pattern recognition, there are different types of classification algorithms used. Here Multi Back Propagation Neural Network [6] is used to classify the images which is segmented, based on the values of feature set.

    Minimum error value is measured by this algorithm. Based on the problem, the number of nodes which is processed and the input layers are chosen. The pattern is given output and the training will be continued until the greater performance is achieved.



    A . Dataset

    An input image is read and it is resized to 250×250, if that image is a coloured one, it should be converted into gray scale image. The range of pixel is between 0 to 255 and the intensity of gray scale image is saved as an 8-bit integer and it

    gives 256 gray shades from black to white. The neural network classifier trains the abnormal image and stores it. The input to be tested is compared with a set of data to detect the plaque. Both normal and abnormal carotid images are present in the test data set. These inputs are tested independently, 38 images are classified correctly and 2 are false negative.

    1. Software Implementation

      The tool which we used to implement this algorithm is MATLAB11a.256×256 sized 40 ultrasound images and 5 normal carotid artery images are tested.

    2. Pre-processing outputs

      Figure 1 Figure 2

      Figure 3 Figure 4

      Figure 1 : Input image

      Figure 2 : Histogram equalized image Figure 3 : Noised image

      Figure 4 : De-noised image

    3. Feature Extraction Outputs

      Moments of Gray-Level Histogram

      2D- Continuous Wavelet Transform

      Gray Level Co-occurrence Matrix

    4. Segmented Outputs

      Figure 1 Figure 2

      Figure 1: Segmented image using SFCM Figure 2: Segmented image using FCMLSM

    5. Classified Outputs


In this work, we have extracted twenty three features from ultrasound images. The features are extracted by Moments of Gray-Level Histogram (MGLH), Gray Level Co- occurrence Matrix (GLCM), Two Dimensional Continuous Wavelet Transform (2D-CWT). It is further given to Genetic Algorithm to remove unnecessary features. Some of the selected features are given as an input to the classifier Multi- Layer Back-Propagation Neural Networks (MLBPNN) which classifies the image whether it is normal or abnormal. Thus our work yields better results than the previous one [7].


  1. R.Polikar, Fundamental Concepts & An Overview of the Wavelet Theory: Tutorial, 1996.

  2. Yongjian Yu and Scott T.Acton Speckle Reducing Anisotropic Diffusion, IEEE Transactions on Image Processing, Vol.11, November 2002.

  3. H.S.Sheshadri, A.Kandaswamy, Experimental investigation on breast tissue classification based on statistical feature extraction of mammograms, Computerized Medical Imaging and Graphics 31(2007) 46-48.

  4. G.Holmes, A.Donkin, I.H.Witten, WEKA: a machine learning workbench, in: Proceedings of the 1994 Second Australian and New Zealand Conference on Intelligent Information Systems, 1994, pp.357- 361.

  5. M.Vasantha, V.Sunniah Bharathi, T.Dhamodharan, Medical image features, extraction, selection and classification, International Journal of Engineering Sciences and Technology 2(2010) 2071-2076.

  6. N.Santhiyakumari, P.Rajendran, M.Madheshwaran, Medical decision- making system of ultrasound carotid artery intima-media thickness using neural networks, Journal of Digital Imaging 24(2011) 1112-1125.

  7. Mehdi Hassana, Asmatullah Chaudhry, Asifullah Khana, Jin Young Kim, Carotid artery image segmentation using modified spatial fuzzy c- means and ensemble clustering, computer medthods and programs in biomedicine 108(2012) 1261-1276.

  8. K.S.Chuang, H.L.Tzeng, S.Chen, J.Wu, T.J.Chen, Fuzzy C-means clustering with spatial information for image segmentation, Computerized Medical Imaging and Graphics: The Official Journal of the Computerized Medical Imaging Society 30(2006)9-15.

Leave a Reply