Analysis of Histopathological Images for Prediction of Breast Cancer Using Traditional Classifiers with Pre-Trained CNN

Abstract

Breast cancer is one of the most commonly found and dangerous cancer among women which leads to a major research topic in medical science. Most of the times, it is identified by using a biopsy method where tissue is removed and studied under a microscope. If a histopathologist is not well trained then this may lead to the wrong diagnosis. In order to facilitate better diagnosis, automatic analysis of histopathology images can help pathologists to identify malignant tumors and cancer subtypes. Recently, Convolutional Neural Networks (CNNs) have become preferred deep learning approaches for breast cancer classification and detection. In this research, two machine learning approaches i.e. Support Vector Machine (SVM) and Logistic Regression (LR) are used for comparative analysis. This paper mainly focuses on leveraging pre-trained (CNN) activation features on traditional classifiers to perform automatic classification of breast cancer images. For this purpose, a two-phase model has been proposed for automatic classification on the basis of magnification subsequently classify the samples for benign and malignant. This model is trained separately with respect to various image magnifications (40x, 100x, 200x and 400x). In this study, the dataset is partitioned into the following fashion: 80% for the training phase and 20% for testing phase The performance is analyzed by using Accuracy, Precision, Recall and F1-score in order to find out the best-suited model that can be used for automation system. The experimental results demonstrate that ResNet50 network has achieved maximum accuracy for LR in comparison to SVM in magnification factor. In addition, results show that the performance of CNN +LR is slightly better than CNN +SVM for classification of benign and malignant classes. The proposed model helps in extracting more accurate image features and significantly improves the overall accuracy of classification.