Biomedical Signal Processing and Control 59 (2020) 101912 Contents lists available at ScienceDirect Biomedical Signal Processing and Control journal homepage: www.elsevier.com/locate/bspc Automated breast cancer detection in digital mammograms: A moth flame optimization based ELM approach Debendra Muduli ∗ , Ratnakar Dash, Banshidhar Majhi Pattern Recognition Laboratory, Department of Computer Science and Engineering, NIT-Rourkela, Odisha 769008, India a r t i c l e i n f o Article history: Received 5 August 2019 Received in revised form 31 December 2019 Accepted 22 February 2020 Keywords: MFO-ELM LWT PCA LDA Mammogram classification CAD model a b s t r a c t Early detection of breast cancer based on a digital mammogram is an important research domain in the field of medical image analysis. An improved CAD model is proposed in this paper for the classification of breast masses into the normal or abnormal and benign or malignant category. The proposed model utilizes lifting wavelet transform (LWT) to extract the features from the region of interest mammogram images. The dimension of the feature vectors is then reduced by using a fusion of PCA and LDA methods. Finally, the classification is performed using a combination of an extreme learning machine and moth flame optimization technique (MFO-ELM). In the MFO-ELM algorithm, MFO is used to optimize the hidden node parameters of ELM. Further, 5-fold stratified cross-validation is used to improve the generalization performance of the model. The proposed model is evaluated on two standard datasets, namely MIAS and DDSM. From the experiment, it is observed that the proposed CAD model obtains ideal results for the MIAS dataset and achieves an accuracy of 99.76% (normal vs. abnormal) and 98.80% (benign vs. malignant) for the DDSM dataset. Our proposed model also demands minimum computational time as compared to other existing models. The experimental results show that the proposed model is superior to other state-of-the-art models in terms of classification accuracy with a significantly reduced number of features. © 2020 Elsevier Ltd. All rights reserved. 1. Introduction In recent years, breast cancer has become one of the most prevalent causes of death among women. According to the report provided by the American Cancer Society, breast cancer cases reach up to 252,710 among women in the US during 2017, and it has been observed that the malignant tumor is the most dangerous. The death rate is expected to 40,610 by 2017. The condition is direr in a less developed country in India. The Globocan project report suggested that breast cancer is common cancer in India and 162,468 new cases were detected every year and the death rate is 87,090 per year [1,2]. Manual detection of screening mammograms by the radio- grapher is tedious, costly, time-consuming and causes a high false-positive rate. However, variation in tissue and lack of exper- tise makes the detection process difficult. To overcome these issues, automated mammogram breast cancer detection systems need to be developed by using dedicated computer systems which can assist radiologists to provide the corrective measures for treating ∗ Corresponding author. E-mail address: muduli.debendra@gmail.com (D. Muduli). the patients at an early stage [3]. In the past decade, several meth- ods have been proposed toward the development of various Hybrid CAD models for mammogram classification [3–7]. Still, the existing systems require better accuracy and lesser computational time to increase the performance of the model and helps the radiographer for diagnosis. Most of the existing mammogram CAD models based on dif- ferent frequency domain transform processes like DWT, DCT, DST for feature extraction [4,8–10]. Wavelet transform has advantages over other transformation methods as preserve spatial information. However, the conventional wavelet transform suffers from com- putational and memory overhead. Hence, in this paper, an efficient wavelet scheme namely lifting wavelet transform (LWT) is used for extracting features from ROIs of mammogram images. The main advantage of LWT is faster computation with less memory space as compared to traditional wavelet transforms. Previously the LWT scheme is used in various works like audio watermarking [11–13], image watermarking [14], human fall detection scheme [15]. How- ever, to the best of our knowledge, its effectiveness has not yet been investigated for the detection of breast cancer in mammogram images. Further, both supervised and unsupervised models have been extensively utilized for classification. Earlier mammogram CAD https://doi.org/10.1016/j.bspc.2020.101912 1746-8094/© 2020 Elsevier Ltd. All rights reserved.