Performance Analysis of Segmentation Algorithms for the Detection of Breast Cancer

Abstract

One of the main reasons for cancer death globally is breast cancer. The detection of breast cancer from Hematoxylin (H) and & Eosin (E) stained pathology image is significant and digital pathologists are struggling to get accurate final decision. The objective of this study was to carry out the performance analysis of segmentation algorithms for the detection of breast cancer. This study proposed a method for computer assisted diagnosis and classification of breast cancer from microscopic slide images by means of biologically explicable features. The suggested methodology consists of different stages involving image enhancement, nuclei segmentation, extraction of features, and to end with the classification. A comparative analysis was done on various segmentation algorithms and finally, a suitable and effective method was utilized in the proposed approach. The contrast-limited adaptive histogram equalization is performed on the input images for contrast enhancement. Similarly, k-means clustering segmentation algorithm is used for segmenting the nuclei in the proposed work because it outperforms the other commonly used algorithms during comparative analysis. Gray level texture features were extracted in the feature extraction step. Finally, support vector machine classifier was employed for classification of breast cancer histopathology images into benign and malignant classes because it is the most efficient classifier comparing to the other usual classifiers for this purpose. The performance analysis of the suggested design was estimated using familiar parameters such as accuracy, sensitivity, specificity, etc. and achieved an accuracy of 91.1 %.