Assessing the Effect of Pre-processing Techniques on Classification of Breast Cancer using Histopathological Images

Abstract

Over past few decades Breast cancer (BC) has become more common and affecting females in early age, which is an alarming and challenging situation for researchers to provide methods to identify the disease in their early stage. This is the deadliest cancer among women and is alarming female fraternity becoming second leading cause of deaths. If the disease gets identified in their early stage it may leads to reduction in mortality rate. It may occur in cells that produce milk (lobules) or in the passages responsible for carrying milk (milk ducts). This paper presents the performance comparison of various pre-processing techniques based on the BreakHis dataset. The dataset used contains 1980 breast histopathological images including 625 benign and 1355 malignant cases. Initially the histopathological images have been pre-processed using techniques including contrast limited adaptive histogram equalization (CLAHE), contrast stretching (CS), histogram equalization (HE), and unsharp masking (UM) followed by feature extraction using 2D Gabor Wavelet Transform to obtain texture feature from both the categories like original and preprocessed images. Finally, support vector machine (SVM) classifies the images in two categories namely benign and malignant. The experiments results show that texture features computed using UM as pre-processing tool outperformed for making difference between benign and malignant images using breast histopathological images with a classification accuracy of 84.1 %.