BREAST CANCER HISTOLOGICAL IMAGES CLASSIFICATION AND PERFORMANCE EVALUATION OF DIFFERENT CLASSIFIERS

Abstract

Breast cancer is a serious trouble and one of the greatest causes of death for women throughout the world. Computer-aided diagnosis (CAD) techniques can help the doctor make more credible decisions. We have determined the possibility of knowledge transfer from natural to histopathological [IX][XII] images by employing a pre-trained network ResNet-50.This pre-trained network has been utilized as a feature generator and extracted features are used to train support vector machine (SVM), random forest, decision tree, and K nearest neighbor(KNN) classifiers[X]. We altered the softmax layer to support the vector machine classifier, random forest classifier, decision tree classifier, and k-nearest neighbor classifier, to evaluate the classifier performance of each algorithm. These approaches are applied for breast cancer classification and evaluate the performance and behavior of different classifiers on a publicly available dataset named Bttheeak-HIS dataset. In order to increase the efficiency of the ResNet[III] model, we preprocessed the data before feeding it to the network. Here we have applied to sharpen filter and data augmentation techniques, which are very popular and effective image pre-processing techniques used in deep models.