Predicting Multi-level Classification of Breast Cancer Image by using Three FCM Variants
Vandana Kate, Pragya Shukla · 2022 International Conference for Advancement in Technology (ICONAT) · 2022
Breast cancer is the second leading cause of death among women, after lung cancer. Its initial realization, on the other hand, may significantly improve the chances of patient survival. Histopathological images of impacted tissues acquired from surgical biopsies provide adequate visual information to differentiate between breast cancer types. Breast Cancer Histopathological Images Analysis (BCHIA) is a widely used method for detecting abnormal pathological alterations in breast cancer. In this paper, we propose a machine learning technique for classifying the cancer cells in two classes and subsequently classifying them into one of four sub classes in each category. In this context the Breast Cancer(BC) dataset with 7780 histopathology images scanned at 40X, 100X, 200X and 400X magnification level is used for examination and early detection of breast cancer. We have used histopathology images as they are considered safe in comparison to other radiology imaging techniques. Additionally three variants of Fuzzy C-Means (FCM) image segmentation techniques are proposed namely GEFCM (Gaussian-Euclidean Fuzzy C-Means), GMFCM (Gaussian-Minkowski Fuzzy C-Means), and GMHFCM (Gaussian-ManHattan Fuzzy C-Means). These techniques are developed with the help of Gaussian kernels RBF (radial basis function) and three different distances measuring functions namely Euclidean distance, Minkowski distance, and ManHattan (Texi-Cab) distance functions. The calculated features using these three techniques of image segmentation are classified into one of BC type using multi-class SVM. The implementation of proposed data model is performed using MATLAB tool. Additionally for justification the accuracy, precision, recall and f1 score is measured. The experimental analysis demonstrates the superiority of GMHFCM (Gaussian-ManHattan Fuzzy C-Means) based extracted edge features.