A Novel Local Binary Patterns-Based Approach and Proposed CNN Model to Diagnose Breast Cancer by Analyzing Histopathology Images

Abstract

Breast cancer, which is among the most common types of cancer in women, is a serious disease that requires attention due to its high risk of death. In this respect, every effort made to help early diagnosis is remarkable. This article proposed two methods, one CNN-based and the other local binary pattern (LBP)-based, to perform the preliminary diagnosis process on breast cancer histopathology images with high performance. Within the scope of the study, the high performance of the two proposed methods was applied to two separate comprehensive datasets most preferred in breast cancer studies. The proposed CNN model has 20 layers and is called quad star LBP (QS-LBP) due to its star-like structure based on the proposed LBP. The histopathology images improved with the QS-LBP method were then analyzed with the most commonly used Random Forest and Optimized Forest algorithms among machine learning algorithms. The BreaKHis dataset contains images with 40X, 100X, 200X, and 400X magnification resolutions and contains approximately 7924 images. The other comprehensive dataset containing analyzed histopathology images contains approximately 278 thousand images. Both datasets examined are extremely important in terms of the significant number of breast cancer histopathology images they contain. The results obtained with the QS-LBP method developed within the scope of the study are 94.58% accuracy, 92.3% F1 score, and 97.9% AUC/ROC, respectively. The results obtained with the proposed CNN model are 98.27% accuracy, 98% F1 score, and 97% AUC/ROC, respectively. The QS-LBP method and CNN model developed within the scope of the study outperform many existing methodologies in classifying breast cancer histopathology images as benign and malignant. In addition to all these, the accuracy of both methods proposed within the scope of the article can be compared with some state-of-the-art methods.