Classifications of Breast Cancer Images by Deep Learning
Wenzhong Liu, Li hualan, Caijian Hua, Liangjun Zhao · medRxiv · 2020
Abstract Background Breast cancer is a leading cause of cancer-related death in women. Classifications of pathological images are important for its diagnosis and prognosis. However, the existing computational methods can sometimes hardly meet the accuracy requirement of clinical applications, due to uneven color distribution and subtle difference in features. Methods In this study, a novel classification method DeepBC was proposed for classifying the pathological images of breast cancer, based on the deep convolution neural networks. DeepBC integrated Inception, ResNet, and AlexNet, extracted features from images, and classified images of benign and malignant tissues. Results Additionally, complex tests were performed on the existing benchmark dataset to evaluate the performance of DeepBC. The evaluation results showed that, DeepBC achieved 92% and 96.43% accuracy rates in classifying patients and images, respectively, with the F1-score of 97.38%, which better than the state-of-the-art methods. Conclusions These findings indicated that, the model had favorable robustness and generalization, and was advantageous in the clinical classifications of breast cancer.