Deep PCANet Framework for the Binary Categorization of Breast Histopathology Images
Revathi Mukkamala, Poreddy Santoshi Neeraja, Sravya Pamidi, Tina Babu, Tripty Singh · 2018
Breast Cancer is one of leading causes for high mortality among women across the world. A computer aided diagnosis tool for examining breast tumor histopathology images and classifying them is in much need. Principal Component Analysis Network (PCANet) is simple and efficient deep learning framework which uses linear approach for feature extraction. In this study, to effectively integrate the color information for learning the features, we propose a framework for analyzing color histopathological images using PCANet. In the presented framework a color histopathology image is split into LAB color space components and cascaded PCA is performed to give principal component images for each component from which color angular pattern and norm pattern images are generated. Simple binary hashing is performed on these color angular and norm pattern images to produce several binary images which are further encoded to extract block-wise histogram features. The extracted features are fed to the SVM classifier for further classification into benign or malignant. Experimental results on BreakHis dataset demonstrates that for breast tumor classification the proposed method shows a significant improvement in performance over the traditional methods. It achieved an accuracy of 97% on 100x magnification which shows its potential as a computer aided diagnosis tool for breast cancer diagnosis.