A novel improved crow‐search algorithm to classify the severity in digital mammograms

• Published in: International journal of imaging systems and technology Vol. 31; no. 2; pp. 921 - 954
• Main Authors: Sannasi Chakravarthy, S R, Rajaguru, Harikumar
• Format: Journal Article
• Language: English
• Published: Hoboken, USA John Wiley & Sons, Inc 01.06.2021; Wiley Subscription Services, Inc
• Subjects: Breast cancer; Classification; crow‐search algorithm and chaotic maps; Datasets; Decomposition; Digital imaging; Discrete Wavelet Transform; Feature extraction; Image analysis; Image classification; mammogram images; Mammography; Medical imaging; Optimization; Optimization algorithms; Randomness; Screening; Search algorithms; wavelet; Wavelet transforms
• ISSN: 0899-9457; 1098-1098
• DOI: 10.1002/ima.22493

The survival rates of breast cancer are going up due to the emerging increase in its screening and diagnosis methods. However, breast cancer is yet the most intrusive disease found in women. Many techniques are emerging during recent years for the investigation of breast cancer using imaging modalities. The paper intends to categorize the severity present in the digital mammography images as either benign (B) or malignant (M) using an improved crow‐search optimization algorithm (ImCSOA). In the literature, the CSOA is generally used for solving several feature selection and numerical optimization problems. The objective is to utilize this popular optimization algorithm for the problem of biomedical image classification. However, if this algorithm is applied directly to classification problems, then it will result in poor classification of data. Hence, the original CSO (OCSO) algorithm undergoes suitable enhancements using a novel controlled parameter tuning, control operator and chaotic‐maps‐based controlled randomness. Four distinct chaotic maps are used for controlling the randomness in the OCSO algorithm. The mammogram images are obtained from the Mammographic Image Analysis Society and Digital Database for Screening Mammography data sets for the evaluation. The classification is accomplished through discrete wavelet transform‐based statistical features that are extracted at two levels [level 4 (L4) and level 6 (L6)] of decomposition. For both data sets, the ImCSOA with L4 and L6 decomposed bior4.4 wavelet features provides the maximum accuracy of around 85% to 86%, which is approximately 62% to 88% better than the OCSO algorithm with L4 and L6 decomposed bior4.4 wavelet features.