Hybrid-based bat optimization with fuzzy C-means algorithm for breast cancer analysis

Abstract

Background: Breast cancer is one of the most frequent types of cancer among women and early identification can reduce the mortality rate drastically. Feature selection is one of the significant tasks in the breast cancer analysis process. Several types of feature selection algorithms have been implemented to select the most appropriate feature for breast cancer analysis. However, they have to take a longer time to converge, over-fitting problems and providing less accuracy. Hence, a hybrid bat optimization algorithm combined with chaotic maps and fuzzy C-means clustering algorithm (BSCFC) is proposed for feature selection. Aims and Objectives: An integrated optimized bat optimization algorithm combined with chaotic maps and fuzzy C-means clustering algorithm (BSCFC) is proposed to determine the relevant feature. Materials and Methods: Breast cancer mini-Mammographic Image Analysis Society database (MIAS) dataset is used for analysis. Further, median filters are used for preprocessing, Region of Interest (ROI) was utilized for segmentation, gray level co-occurrence matrix (GLCM), and texture analysis are utilized in the feature extraction process. A hybrid bat optimization algorithm combined with chaotic maps and fuzzy C-means clustering algorithm (BSCFC) is proposed for feature selection. K nearest neighbor (KNN) classifier is used for classification. Results: Performance of the proposed system is evaluated using standard measures and achieved the highest accuracy rate of (98.2%), specificity of (97.3%), and sensitivity of (98.3%) as compared to other relevant methods such as bat, chaotic bat, chaotic crow search, ant lion optimization, and chaotic ant lion optimization algorithm. Conclusion: The proposed BSCFC algorithm is designed to improve the performance of convergence speed and control balance between exploration and exploitation rate using five types of chaotic maps namely sinusoidal, sine, gauss, logistic, and tent maps. The results show that the BSCFC with sinusoidal maps can significantly boost the classification performance of the BSCFC algorithm in classifying the breast cancer images with reduced features, which in turn optimizes the radiologists' time for their interpretation.