Classification of Breast Cancer Histopathological Image Based on Lightweight Network

Inhalt:
Breast cancer is a common cancer in female cancer patients, and histopathological image analysis is the gold standard for cancer diagnosis. Therefore, it is of great significance to use computer-aided diagnosis for automatic classification of breast cancer pathological images. Research in this direction mainly focuses on improving the recognition rate of breast cancer pathological images. This paper proposes a lightweight convolutional neural network named DgNet for the problem of network parameters and flops. The feature maps extracted from the existing convolutional neural networks have redundancy, and these redundancy can be obtained by linearly transforming the feature maps. This linear transformation of this feature map reduces the amount of parameters and flops of the network relative to the convolution operation. In addition, this paper also enhances and normalizes the data set to improve the recognition rate of the network for breast cancer pathological images. In the public dataset BreaKHis, the network in this paper achieves an average recognition rate of 95.87% for the benign and malignant binary classification of breast cancer pathological images. Experiments show that this paper achieves better results with fewer parameters and flops.