Elsevier

IRBM

Volume 43, Issue 1, February 2022, Pages 49-61
IRBM

Original Article
Automatic Detection and Classification of Mammograms Using Improved Extreme Learning Machine with Deep Learning

https://doi.org/10.1016/j.irbm.2020.12.004Get rights and content

Highlights

  • A classification framework is proposed to solve three-class breast cancer problem.

  • Deep-learning based robust features are extracted using pretrained ResNet18 model.

  • Extreme Learning Machine is employed for the detection and classification problem.

  • The performance is further enhanced using the proposed optimization technique.

Abstract

Background and objective

Breast cancer, the most intrusive form of cancer affecting women globally. Next to lung cancer, breast cancer is the one that provides a greater number of cancer deaths among women. In recent times, several intelligent methodologies were come into existence for building an effective detection and classification of such noxious type of cancer. For further improving the rate of early diagnosis and for increasing the life span of victims, optimistic light of research is essential in breast cancer classification. Accordingly, a new customized method of integrating the concept of deep learning with the extreme learning machine (ELM), which is optimized using a simple crow-search algorithm (ICS-ELM). Thus, to enhance the state-of-the-art workings, an improved deep feature-based crow-search optimized extreme learning machine is proposed for addressing the health-care problem. The paper pours a light-of-research on detecting the input mammograms as either normal or abnormal. Subsequently, it focuses on further classifying the type of abnormal severities i.e., benign type or malignant.

Materials and methods

The digital mammograms for this work are taken from the Curated Breast Imaging Subset of DDSM (CBIS-DDSM), Mammographic Image Analysis Society (MIAS), and INbreast datasets. Herein, the work employs 570 digital mammograms (250 normal, 200 benign and 120 malignant cases) from CBIS-DDSM dataset, 322 digital mammograms (207 normal, 64 benign and 51 malignant cases) from MIAS database and 179 full-field digital mammograms (66 normal, 56 benign and 57 malignant cases) from INbreast dataset for its evaluation. The work utilizes ResNet-18 based deep extracted features with proposed Improved Crow-Search Optimized Extreme Learning Machine (ICS-ELM) algorithm.

Results

The proposed work is finally compared with the existing Support Vector Machines (RBF kernel), ELM, particle swarm optimization (PSO) optimized ELM, and crow-search optimized ELM, where the maximum overall classification accuracy is obtained for the proposed method with 97.193% for DDSM, 98.137% for MIAS and 98.266% for INbreast datasets, respectively.

Conclusion

The obtained results reveal that the proposed Computer-Aided-Diagnosis (CAD) tool is robust for the automatic detection and classification of breast cancer.

Introduction

Breast cancer remains as a deadly cancer among women and globally, it is the second prominent cancer. The rate of mortality of breast cancer is substantially very high than others [1], next to lung cancer. The occurrence rate of breast cancer has been increased progressively in both economically developing as well in developed nations. Furthermore, 15% of deaths resulting from breast cancer are found because of the lack of earlier detection [2]. The exact root-cause of breast cancer does not have any proper history, which makes the early detection of breast cancer to fall on the dark spot. The challenges in the identification of such cancer rely on the dense and uneven structure of women's breasts. Anyhow, the early detection of breast cancer typically elevates the life span of the affected victims [3]. To strengthen the survival rate, an efficient diagnostic methodology is vital for the timely detection of such cancer.

Various imaging techniques are adopted for imaging the breast for earlier identification and diagnosis. Among those techniques, ultrasound imaging modality is popular because it uses acoustic waves (zero ionizing radiation) to image the women's breasts. Although it uses zero radiation, it lacks in providing detailed information such as microcalcification [4]. But the timely detection of such cancer primarily requires detailed information about microcalcification. The microcalcification in breast cells is represented as a minimal or tiny deposit of calcium that is very hard to be felt through any symptoms but can be easily recognized using imaging modalities.

Mammography, a specialized and non-invasive imaging modality dedicated for imaging the breast. Mammography is widely used since it adopts lower radiation of x-ray to recognize the microcalcifications found in the breast. The prime advantage of mammogram imaging modality is that it could recognize cancer even before a victim can sense the symptoms physically [4]. Hence, digital mammogram images obtained using mammography influence the possibility of early detection of breast cancer. The manual reading of acquired mammograms is one of the most tedious tasks for radiologists. The manual analysis of mammogram images with naked eyes will always result in an incorrect diagnosis. It might sound better if the microcalcification present in mammogram images is detected and its severities are categorized automatically by using a Computer-Aided Diagnosis (CAD) system. The knowledge of experience as well as the design of such computer-aided tools will further improve the classification accuracy of any model. The severity of microcalcification denotes the nature of abnormality found within the breast, which belongs to either benign type or malignant type of severity [5]. Here, the benign type of severity is generally non-cancerous i.e. the benign tumors will not invade to its neighbor tissues. But the malignant type of severity is generally claimed as a cancerous one because the malignant tumors often invade aggressively and capable of spreading to the neighbor tissues of the breast. Thus, it is required to detect and distinguish between the types of severities that possibly enhances the earlier diagnosis of breast cancer.

Fig. 1 illustrates the proposed workflow for the problem of detection and classification of breast cancer using digital mammogram images. After pre-processing, the features are extracted using ResNet18 deep-learning architecture, and then these features are normalized. Subsequently, feature analysis is carried out for ensuring the suitable type of classifier for the problem. Then, the proposed Improved Crow-Search Optimized Extreme Learning Machine (ICS-ELM) algorithm is implemented for the detection and classification of microcalcification in digital mammogram images. The prime goal of the proposed work is to detect breast cancer and then classifying the severity of breast cancer (benign or malignant severity) effectively. The paper uses MATLAB R2020a installed on Windows-10 professional OS with Intel i3 processor having 4 GB of RAM and one TB of hard disk memory for its implementation.

The work is structured as Section 2 discusses the motivation and related works. Section 3 describes the materials and techniques for feature extraction and feature analysis. The existing ELM and crow-search algorithm are summarized in Section 4. The proposed method for this three-class classification is illustrated in Section 5. The outcome of the proposed work is analyzed (individual and overall) in Section 6. Subsequently, Section 7 will finally conclude the paper.

Several research studies [6], [7], [8], [9] have been proposed recently for the design of deep learning-based mammogram classification frameworks, but the work intends to utilize the deep learning-based extraction of features and aims to improve the classification of ELM network with the use of a simple metaheuristic algorithm.

Section snippets

Related works

In the last two decades of research towards the diagnosis of breast cancer, the community of medical imaging has been delivering several promising approaches to develop improved and effective CAD tools. In this regard, various researchers have proposed several techniques for feature extraction with appropriate improvements in the detection and classification part.

Support Vector Machine [9] shortly known as SVM is a popular algorithm employed in various machine-learning tasks. This is because of

Datasets used

The proposed work is evaluated using digital mammogram images as input downloaded from the CBIS-Digital Database for Screening Mammography (DDSM), Mammographic Image Analysis Society (MIAS), and INbreast database.

The digital mammograms in the DDSM dataset are created and retained by the University of South Florida [20]. The DDSM contains digital mammograms that are acquired from approximately 2500 cases with 43 volumes. Moreover, four digital mammograms are acquired with both views:

Extreme Learning Machine (ELM)

ELM, proposed by Huang et al. [12] in 2006, is a faster training algorithm used in SLFN networks. As given in Fig. 3, the SLFN architecture is just a neural network with one hidden layer that interconnects two layers (input with output layer). In the ELM, the input weights are the weights calculated between input nodes and hidden neurons. Herein, the hidden biases are used for interconnecting all the hidden neurons whereas the output weights are used for interconnecting the hidden neurons to

Proposed ICS-ELM algorithm

In most of the metaheuristic optimization techniques, a few parameters are needed to be randomly assigned either with uniform or Gaussian distribution. This results in a slower convergence rate or inconsistency in the process of optimization. Hence, to overcome this problem, necessarily two modifications in CSOA are done to enhance its rate of convergence: the first one is the inclusion of a simple control parameter which makes the crows in the flock to search for global minima, consequently

Results and discussion

To examine the effectiveness of the proposed ICS-ELM, the standard SVM with RBF kernel (SVM-RBF) [36], ELM [37], PSO-ELM [38], and Chaotic Crow-Search ELM (CS-ELM) [34] are used. The employed kernel in SVM is RBF kernel because the input considered for our problem is highly non-linear (as shown in Fig. 2). The ICS-ELM adopts two distinct chaotic map functions: logistic map with ICS-ELM (ICS-ELM1) and sine map with ICS-ELM (ICS-ELM2). Hence, the work examines the performance comparison of

Conclusion

For decreasing the cancer deaths among women due to breast cancer, early detection with an effective classification framework is always required in healthcare applications. This paper proposed a promising classification methodology that utilizes deep methods (convolutional neural network), classification robustness of extreme learning machine, chaotic maps, and better searching-capability of the crow-search optimization algorithm. The work used ResNet18 architecture for extracting the feature

Human and animal rights

The authors declare that the work described has not involved experimentation on humans or animals.

Funding

This work did not receive any grant from funding agencies in the public, commercial, or not-for-profit sectors.

Author contributions

All authors attest that they meet the current International Committee of Medical Journal Editors (ICMJE) criteria for Authorship.

Declaration of Competing Interest

The authors declare that they have no known competing financial or personal relationships that could be viewed as influencing the work reported in this paper.

References (51)

  • P.K. Wong et al.

    Modeling and optimization of biodiesel engine performance using kernel-based extreme learning machine and cuckoo search

    Renew Energy

    (2015)
  • M. Luo et al.

    Compound feature selection and parameter optimization of ELM for fault diagnosis of rolling element bearings

    ISA Trans

    (2016)
  • C. Zhang et al.

    A compound structure of ELM based on feature selection and parameter optimization using hybrid backtracking search algorithm for wind speed forecasting

    Energy Convers Manag

    (2017)
  • W. Fu et al.

    Multi-step short-term wind speed forecasting approach based on multi-scale dominant ingredient chaotic analysis, improved hybrid GWO-SCA optimization and ELM

    Energy Convers Manag

    (2019)
  • I.C. Moreira et al.

    Inbreast: toward a full-field digital mammographic database

    Acad Radiol

    (2012)
  • S. Sreedevi et al.

    A novel approach for removal of pectoral muscles in digital mammogram

    Proc Comput Sci

    (2015)
  • Z. Wang et al.

    Breast tumor detection in digital mammography based on extreme learning machine

    Neurocomputing

    (2014)
  • R. Taormina et al.

    Data-driven input variable selection for rainfall–runoff modeling using binary-coded particle swarm optimization and extreme learning machines

    J Hydrol

    (2015)
  • Z. Hua et al.

    2D sine logistic modulation map for image encryption

    Inf Sci

    (2015)
  • I. Buciu et al.

    Directional features for automatic tumor classification of mammogram images

    Biomed Signal Process Control

    (2011)
  • P. Görgel et al.

    Mammographical mass detection and classification using local seed region growing–spherical wavelet transform (lsrg–swt) hybrid scheme

    Comput Biol Med

    (2013)
  • A.S. Setiawan et al.

    Mammogram classification using law's texture energy measure and neural networks

    Proc Comput Sci

    (2015)
  • M. Abdel-Nasser et al.

    Analysis of tissue abnormality and breast density in mammographic images using a uniform local directional pattern

    Expert Syst Appl

    (2015)
  • U. Raghavendra et al.

    Application of Gabor wavelet and locality sensitive discriminant analysis for automated identification of breast cancer using digitized mammogram images

    Appl Soft Comput

    (2016)
  • K. Hu et al.

    Microcalcification diagnosis in digital mammography using extreme learning machine based on hidden Markov tree model of dual-tree complex wavelet transform

    Expert Syst Appl

    (2017)
  • Cited by (64)

    • Breast cancer diagnosis: A systematic review

      2024, Biocybernetics and Biomedical Engineering
    • Edge detection and graph neural networks to classify mammograms: A case study with a dataset from Vietnamese patients

      2023, Applied Soft Computing
      Citation Excerpt :

      The prediction performance of the OMLTS-DLCN model achieved higher accuracy of 98.50% and 97.55% on the Mini-MIAS dataset and DDSM dataset, respectively. Chakravarthy et al. [51] presented a new customized method of integrating the concept of deep learning with the extreme learning machine (ELM), which was optimized using a simple crow-search algorithm (ICS-ELM). The algorithm focused on detecting the input mammograms as either normal or abnormal.

    View all citing articles on Scopus
    View full text