An improved scheme for digital mammogram classification using weighted chaotic salp swarm algorithm-based kernel extreme learning machine

Over the past years, the surge in the necessity for early detection and diagnosis of breast cancer has resulted in many innovative research directions. According to the World Health Organization, an early and accurate detection of breast cancer successfully leads to a correct decision towards its treatment. Development of computer-aided diagnosis (CAD) system is considered to be a major stead in current research practice to abet medical practitioners in decision-making. This paper proposes an improved CAD framework to correctly classify the digital mammograms into normal or abnormal, and further, benign or malignant. The proposed scheme employs a block-based discrete wavelet packet transform (BDWPT) to extract the features, namely, the Shannon entropy, Tsallis entropy, Renyi entropy, and energy. Then, principal component analysis (PCA) technique is utilized to extract the discriminating features from the original feature vector. Subsequently, an optimized wrapper-based kernel extreme learning machine (KELM) using a weighted chaotic salp swarm algorithm (WC-SSA) is proposed as classifier to classify the digital mammograms. Since the efficacy of KELM algorithm depends on its two important parameters, namely, the penalty parameter and the kernel parameter, the prime objective of the proposed work is to obtain the optimized value of the aforementioned parameters as well as to select the most relevant features from the reduced feature vector simultaneously. The proposed scheme is evaluated on three publicly available standard datasets, namely, MIAS, DDSM, and BCDR to validate the efficacy of the proposed BDWPT+PCA+WC-SSA-KELM scheme. The performance of the proposed model is evaluated in terms of different metrics, namely, classification accuracy, sensitivity, specificity, area under curve (AUC), Matthew’s correlation coefficient (MCC), and F-measure via a 5 × 5 stratified cross-validation approach. From the experimental results and their analysis, it is observed that for the normal–abnormal category, the proposed technique results in an accuracy of 99.62% and 99.92% for MIAS and DDSM, respectively, whereas in the case of benign–malignant classification, the proposed method yields an accuracy of 99.28%, 99.63%, 99.60% for MIAS, DDSM, and BCDR datasets, respectively. Further, it is also observed that the proposed WC-SSA-KELM scheme exhibits superior performance as compared to that of its counterparts. Additionally, two well-known statistical analyses, namely, ANO