A Hybrid Method Based on Feature Fusion for Breast Cancer Classification using Histopathological Images

Abstract

Breast cancer is the most common type of cancer in women today, and it ranks second after lung cancer with a very high mortality rate. If it is detected late, the treatment of breast cancer becomes very difficult. Although there are various methods for the detection of breast cancer, there is still a need for auxiliary diagnosis and treatment methods. In this study, a hybrid method is proposed to investigate the development of basal-like breast tumors and classify basal-like breast cancer types using histopathological images. In the study, firstly, appropriate features that support the accurate classification between tumor and non-tumor regions are extracted from histopathological images. Then the dataset is created by combining the obtained features. In the last stage of the study, the classification of images is carried out by using bag of words (BoW) and deep neural networks (DNN) techniques in a hybrid manner. Generally, immunohistochemical markers are used for this classification, but the performance of these markers remains at 60%. The performance of the classification accuracy of the proposed system is increased with the proposed hybrid classifier based on feature fusion. As a result of the study, 94.5% classification accuracy is achieved on the training set, while 80.8% classification accuracy is succeed on the test set. As a result, it is verified that successful results are achieved in the classification of basal-like breast cancer on histopathological images using the proposed hybrid method based on feature fusion.