Breast Cancer Image Classification via Multi-level Dual-network Features and Sparse Multi-Relation Regularized Learning

Breast cancer is one of the leading causes of cancer death worldwide. Recently, the computer-aided diagnosis and detection technique has been developed for the early diagnosis of breast cancer, but the diagnostic efficiency has still been a challenging issue. For this reason, we aim to improve the breast cancer diagnostic accuracy and reduce the workload of doctors in this paper by devising a deep learning framework based on histological image. Therefore, we develop a model of multi-level feature of dual-network combined with sparse multi-relation regularized learning method, which enhances the classification performance and robustness. Specifically, first, we preprocess the histological images using scale transformation and color enhancement methods. Second, the multi-level features are extracted from preprocessed images using Inception V3-ML and ResNet-50 networks. Third, the feature selection method via sparse multi-relation regularization is further developed for performance boosting and overfitting reduction. We evaluate the proposed method based on the public ICIAR 2018 Challenge dataset of breast cancer histology images. Experimental results show that our method has achieved promising performance and outperformed the related works.