Breast Cancer Image Classification via Multi-Network Features and Dual-Network Orthogonal Low-Rank Learning
Summary
Histopathological image analysis is an important means of clinical early diagnosis and detection of breast cancer. However, its efficiency is limited, so breast cancer detection remains an open problem in medical image analysis. In order to improve the accuracy of early diagnosis of breast cancer and reduce the workload of doctors, this paper combines deep learning and machine learning methods to design a classification framework based on histological images. Specifically, we design a multi-network feature extraction model using a pre-trained deep convolutional neural network (DCNN), develop an effective feature dimensionality reduction method, and train an ensemble support vector machine (E-SVM ) . First, we preprocess the histological images by scaling and color enhancement methods. Second, four pre-trained DCNNs (e.g., DenseNet-121, ResNet-50, multi-stage sink v3, and multi-stage VGG-16) are used to extract multi-network features. Third, a dual-network orthogonal low-rank learning (DOLL) based feature selection method is further developed to improve performance and alleviate overfitting. Finally, the E-SVM is trained to perform a classification task by fusing features and voting strategies to classify images into four categories (i.e., benign, carcinoma in situ, invasive carcinoma, and normal). We evaluated the proposed method on the publicly available ICIAR 2018 Breast Cancer Histology Image Challenge dataset and achieved a high classification accuracy of 97.70%. Experimental results show that our method can achieve good performance and outperform existing methods.
INDEX TER