Contourlet-based mammography mass classification using the SVM family

https://doi.org/10.1016/j.compbiomed.2009.12.006Get rights and content

Abstract

This paper is concerned with the design and development of an automatic mass classification of mammograms. The proposed method consists of three stages. In the first stage, preprocessing is performed to remove the pectoral muscles and to segment regions of interest. In the next stage contourlet transform is employed as a feature extractor to obtain the contourlet coefficients. This stage is completed by feature selection based on the genetic algorithm, resulting in a more compact and discriminative texture feature set. This improves the accuracy and robustness of the subsequent classifiers. In the final stage, classification is performed based on successive enhancement learning (SEL) weighted SVM, support vector-based fuzzy neural network (SVFNN), and kernel SVM.

The proposed approach is applied to the Mammograms Image Analysis Society dataset (MIAS) and classification accuracies of 96.6%, 91.5% and 82.1% are determined over an efficient computational time by SEL weighted SVM, SVFNN and kernel SVM, respectively. Experimental results illustrate that the contourlet-based feature extraction in conjunction with the state-of-art classifiers construct a powerful, efficient and practical approach for automatic mass classification of mammograms.

Introduction

Breast cancer is reported as the second most deadly cancer in the world, on which public awareness has been increasing during the last few decades [1]. Early detection can play an effective role in prevention, particularly by the most reliable detection technology known as mammography.

At the early stages of breast cancer, the clinical signs are very mild and vary in appearance, making diagnosis difficult even for specialists. Therefore, automatic reading of medical images becomes highly desirable and to some extents inevitable. It has been proven that double reading of a mammogram, by two radiologists, reduces missed detection rate, but at a considerable expense. Consequently, the motivation for computer-aided diagnosis is to assist medical staff in achieving higher efficiency and higher accuracy.

Our objective is to develop an automated imaging system for mass classification of digital mammograms. Mass classification requires a preprocessing step of segmenting the input image into disjoint areas, such as the breast region, background, and redundant labels and content text. A significant body of research has already been devoted to breast segmentation including adaptive thresholding [2], [3], statistical approaches, region growing, polynomial modeling [4], active contours [5], template matching, fuzzy technique, multiscale methods, and classifier-based techniques [6], [1]. A summary of different segmentation methods is described in Table 1. Also the reader is referred to [7] for detailed review.

The second step in mass classification (as the most effective stage) is feature extraction. Texture is a commonly used feature in the analysis and interpretation of images. Here, we suggest to distinguish underlying textures in mammography based on Malagelada's approach [4] in the three following stages:

  • 1.

    Statistical methods: The extracted features of this class include those obtained from co-occurrence matrices, surface variation measurements (smoothness, coarseness and regularity) [4], and run-length statistics [6], [8].

  • 2.

    Model-based methods: The analysis of texture features in this class is based on prior models such as Markov random fields [9], auto-regressive models, and fractals [4].

  • 3.

    Signal processing methods: In this class, texture features are obtained according to either pixel characteristics or image frequency spectrum including Laws energy filtering [4], Gabor filtering [4] and wavelets [10], [11], [12].

Following feature extraction, an appropriate classifier is utilized in breast mass classification. Various methodologies have been proposed with the most efficient ones known as Bayesian classifier [13], multilayer perceptron [9], [14], adaptive nero fuzzy inference system (ANFIS) classifier [12], radial basis function (RBF) [14], k-nearest neighbors (KNN) [8], decision tree classifier [15], and support vector machines (SVM) [1]. Table 2 lists advantages and disadvantages of the state-of-the-art mass detection classifiers.

This work is aimed at improving performance of the current mass classification methods using the recently proposed image transforms and classifiers. The novelty of this research is in exploiting the superiority of contourlets [16] in representing line singularities over wavelets. Furthermore the structural risk minimization property of successive enhancement learning (SEL) weighted SVM, support vector-based fuzzy neural network (SVFNN), and kernel SVM achieve a more efficient mammogram mass classification [17].

The rest of this paper is organized as follows: Section 2 gives a background review, Section 3 introduces our proposed methodology and experimental results are described in Section 4. Section 5 will bring this paper to the final step by making conclusions and offering future research directions.

Section snippets

Contourlet transform

The practical approaches for image representation are known as adaptive and non-adaptive representations [18].

As an adaptive representation, Lagrangian method is constructed using full knowledge of intrinsic in image representation. This could mean that edge curves are a priori known, and one constructs an image representation adapted to the structure of those curves [19].

On the other hand, Eulerian representation is fixed, i.e., is a non-adaptive approach, constructed once and for all and has

Methodology

As mentioned above, the proposed system is developed upon the contourlet analysis to extract features and based on the SVM family to classify the breast abnormalities. The proposed system consists of three main stages as shown in Fig. 4(a): segmentation, feature extraction and reduction, and classification processes. Next, we explain and discuss each step separately.

Results and discussions

This section presents and evaluates results of the experiments carried out according to three stages of mammogram examination introduced in this article: (1) image segmentation, (2) ROI feature extraction and reduction and (3) mass classification.

The method was applied to a set of 90 (60 normal and 30 abnormal cases) MLO mammograms taken from the MIAS dataset. The images were in 8-bit gray resolution format and of size 1024 ×1024 pixels.

First, pectoral muscle was successfully separated from the

Conclusions

In this paper, we exploited the advantages of contourlet-based texture analysis along with the geometrical and statistical features and employed SEL weighted SVM, SVFNN and kernel SVM classifiers to detect and classify the breast masses into benign and malignant cases.

It was shown that contourlet transform can successfully capture structural information along multiple scales, locations and orientations. This representation offers improvements over the separable 2-D wavelet transform which has

Conflict of interest statement

None declared.

Fatemeh Moayedi was born in Shiraz, Iran, in May 1982. She received her B.S. degree in hardware engineering from Shiraz University, Shiraz, Iran, in September 2004, where she was honored as the first rank student. She received her M.S. degree from Shiraz University, Shiraz, Iran, in June 2007, where she submitted her thesis “Mammography mass detection and classification”. Since 2008, she has started her Ph.D. in artificial intelligence at Shiraz University. Her research interest includes

References (26)

  • L. Semler, L. Dettori, J. Furst, Wavelet-based texture classification of tissues in computed tomography, in: IEEE...
  • L. Semler, L. Dettori, A comparison of wavelet-based and ridgelet-based texture classification of tissues in computed...
  • E.A. Fischer, J.Y. Lo, M.K. Markey, Bayesian networks of BI-RADS descriptors for breast lesion classification, in: IEEE...
  • Cited by (111)

    • Feature fusion and Ensemble learning-based CNN model for mammographic image classification

      2022, Journal of King Saud University - Computer and Information Sciences
    • Fast discrete curvelet transform and modified PSO based improved evolutionary extreme learning machine for breast cancer detection

      2021, Biomedical Signal Processing and Control
      Citation Excerpt :

      In mammogram classification, feature extraction and selection are the important phases that affect the mammogram classification accuracy performance. Most of the CAD models use several features exploiting like spatial domain, frequency domain, textural properties, statistical properties, multi-resolution behavior, etc., to detect and classify the abnormalities of mammogram images [3,15,16]. The multi-resolution schemes like wavelet and curvelet provide an efficient sparse representation of an image, and [15] several CAD models utilize wavelet features for classification problems.

    View all citing articles on Scopus

    Fatemeh Moayedi was born in Shiraz, Iran, in May 1982. She received her B.S. degree in hardware engineering from Shiraz University, Shiraz, Iran, in September 2004, where she was honored as the first rank student. She received her M.S. degree from Shiraz University, Shiraz, Iran, in June 2007, where she submitted her thesis “Mammography mass detection and classification”. Since 2008, she has started her Ph.D. in artificial intelligence at Shiraz University. Her research interest includes computer vision, statistical image processing, visual tracking, pattern recognition and machine learning.

    Zohreh Azimifar received B.Sc. degree in computer science and engineering from Shiraz University, Shiraz, Fars, Iran, in 1994 and the Ph.D. degree in systems design engineering from the University of Waterloo, Waterloo, ON, Canada, in 2005. In 2005, she was a Postdoctorate Fellow in Medical Biophysics at the University of Toronto, Toronto, ON, Canada. Since 2006, she has been a Faculty Member in Computer Science and Engineering at Shiraz University. Her research interest includes AI in games as well as statistical learning of predictive models for massive data sources with particular attention in computer vision and pattern recognition.

    Reza Boostani was born on 1973, he received his B.Sc. in Electronics from Shiraz University, Shiraz, Iran, in 1996 and his M.Sc. and Ph.D. in Biomedical Engineering from Amirkabir University of Technology, Tehran, Iran, in 1999 and 2004, respectively. He has spent his research period in Graz University of Technology in the BCI field from 2002 to 2003. Since 2004, he has been a faculty member of the Computer Science and Engineering Department of Shiraz University. His current research interests include biomedical signal processing, statistical pattern recognition and machine learning.

    Seraj D. Katebi received the Graduate (Hons.) degree in computer systems engineering from Coventry University, Coventry, UK, in 1972, and the M.Sc. and Ph.D. degrees in automatic control from the Control Systems Center, University of Manchester Institute of Science and Technology (UMIST), Manchester, UK, in 1973 and 1976, respectively. In 1976, he joined the Department of Computer Science and Engineering, Shiraz University, Shiraz, Iran, where he has been a Full Professor since 1993, and is engaged in teaching undergraduate and graduate courses and conducting research in various aspects of nonlinear control and artificial intelligence (AI). He is the author or coauthor of several papers published in various international journals.

    View full text