Breast cancer classification in pathological images based on hybrid features

Breast cancer has become an important factor affecting human health. Diagnosis based on pathological images is considered the gold standard in the clinic. In this paper, an automatic breast cancer detection method based on hybrid features is proposed for pathological images. To obtain better segmentation results under conditions of crowded and chromatin-sparse nuclei, a 3-output convolutional neural network (CNN) is employed to segment the nuclei. Due to the weak correlation between the hematoxylin (H) and eosin (E) channels, texture features are separately extracted for the two channels, which provides more representative results. From multiple perspectives, the morphological features, spatial structural features and texture features are extracted and fused. Using a support vector machine (SVM) classifier with improved generalization, the pathological image is classified as benign or malignant on the basis of the relief method for feature selection. For the University of California, Santa Barbara database (UCSB), the classification accuracy of the method is 96.7%, and the area under the curve (AUC) is 0.983. The experimental results show that the proposed method yields superior classification performance compared with existing techniques.