Publication

Abstract : Breast cancer appears to be the most dangerous disease infecting women. Invasive ductal carcinoma is the most prevalent type of cancer that affects the breast. One in eight ladies suffer from carcinoma and even it leads to the removal of their breast. This study describes a hybrid feature-based approach for detecting ınvasive ductal carcinoma in WSI images. An optimal set of hybrid features is extracted from the images using deep learning models and handcrafted feature extraction techniques. In this work, we have done a comparison of feature sets considering the combination of handcrafted features like grey level co-occurrence matrix, Gabor filter and linear binary pattern with different deep learning models. A set consists of 9 handcrafted and 1000 deep learning features that are combined and used to classify the image patches as IDC or non-IDC by using a k-NN classifier. The experiment was performed on a Kaggle dataset, having breast histopathology images with 277,524 patches. Experimental results have shown that the handcrafted features with the MobileNet model features gave better classification accuracy, F1-score, precision and recall (91%, 87%, 88% and 85%, respectively).