Original ArticleAutomatic Detection and Classification of Mammograms Using Improved Extreme Learning Machine with Deep Learning
Graphical abstract
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)
- et al.
Breast mass detection in digital mammogram based on gestalt psychology
J Healthc Eng
(2018) - et al.
Mammogram classification using law's texture energy measure and neural networks
Proc Comput Sci
(2015) - et al.
Simultaneous detection and classification of breast masses in digital mammograms via a deep learning YOLO-based CAD system
Comput Methods Programs Biomed
(2018) - et al.
A fully integrated computer-aided diagnosis system for digital X-ray mammograms via deep learning detection, segmentation, and classification
Int J Med Inform
(2018) - et al.
Evaluation of deep learning detection and classification towards computer-aided diagnosis of breast lesions in digital X-ray mammograms
Comput Methods Programs Biomed
(2020) - et al.
Basic concepts of artificial neural network (ANN) modeling and its application in pharmaceutical research
J Pharm Biomed Anal
(2000) - et al.
A hybrid algorithm for artificial neural network training
Eng Appl Artif Intell
(2013) - et al.
Extreme learning machine: theory and applications
Neurocomputing
(2006) - et al.
Extreme learning machine and adaptive sparse representation for image classification
Neural Netw
(2016) - et al.
An integrated PSO for parameter determination and feature selection of ELM and its application in classification of power system disturbances
Appl Soft Comput
(2015)
Modeling and optimization of biodiesel engine performance using kernel-based extreme learning machine and cuckoo search
Renew Energy
Compound feature selection and parameter optimization of ELM for fault diagnosis of rolling element bearings
ISA Trans
A compound structure of ELM based on feature selection and parameter optimization using hybrid backtracking search algorithm for wind speed forecasting
Energy Convers Manag
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
Inbreast: toward a full-field digital mammographic database
Acad Radiol
A novel approach for removal of pectoral muscles in digital mammogram
Proc Comput Sci
Breast tumor detection in digital mammography based on extreme learning machine
Neurocomputing
Data-driven input variable selection for rainfall–runoff modeling using binary-coded particle swarm optimization and extreme learning machines
J Hydrol
2D sine logistic modulation map for image encryption
Inf Sci
Directional features for automatic tumor classification of mammogram images
Biomed Signal Process Control
Mammographical mass detection and classification using local seed region growing–spherical wavelet transform (lsrg–swt) hybrid scheme
Comput Biol Med
Mammogram classification using law's texture energy measure and neural networks
Proc Comput Sci
Analysis of tissue abnormality and breast density in mammographic images using a uniform local directional pattern
Expert Syst Appl
Application of Gabor wavelet and locality sensitive discriminant analysis for automated identification of breast cancer using digitized mammogram images
Appl Soft Comput
Microcalcification diagnosis in digital mammography using extreme learning machine based on hidden Markov tree model of dual-tree complex wavelet transform
Expert Syst Appl
Cited by (64)
Breast cancer diagnosis: A systematic review
2024, Biocybernetics and Biomedical EngineeringThe power of deep learning for intelligent tumor classification systems: A review
2023, Computers and Electrical EngineeringA BERT encoding with Recurrent Neural Network and Long-Short Term Memory for breast cancer image classification
2023, Decision Analytics JournalEdge detection and graph neural networks to classify mammograms: A case study with a dataset from Vietnamese patients
2023, Applied Soft ComputingCitation 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.