Deep Learning Extraction of Features for Early Detection of Breast Cancer in Histological Images

Abstract

The Deep Neural Features Extraction is new method of extracting information from an image, this method focuses on extracting the portion of an image consisting of mitosis in histological images, which need to be segmented and processed for detection of breast cancer in human beings. Thus, we need a method that identifies the region of histology images having mitosis. As these mitosis is a sign of early stages of breast cancer, we proposed an algorithm for detecting the region of images through deep neural features. These histology images are preprocessed for enhancing the region that represents the mitosis, as these mitosis are primary cause of the breast cancer, we incorporated a new method of Deep Neural Features that may be helpful in extracting the required information from an image at different layers. Each layer contributes sufficient information towards detecting the mitosis in histology images, thereby the cancerous tissues are detected and stages of breast cancer is determined, whether the stages of image is benign or malignant. The proposed deep neural feature has yielded an accuracy of 98.38 % over a benchmark dataset.