基于Cycle-GAN和改进DPN网络的乳腺癌病理图像分类

针对病理图像染色不均匀及良恶性难以鉴别的问题，提出基于Cycle-GAN和改进的双路径网络（DPN）的算法框架. 利用Cycle-GAN进行颜色归一化处理，解决因病理图像染色不均匀导致的检测模型精度偏低问题，通过对图像进行重叠切片，基于DPN网络采用增加小卷积、反卷积和注意力机制，增强模型对病理图像纹理特征的分类能力. 在BreaKHis数据集上的实验结果表明，所提算法有效提高了乳腺癌病理图像良恶性分类的准确性.

Breast cancer pathological image classification based on Cycle-GAN and improved DPN network

An algorithm framework based on cycle-consistent adversarial network (Cycle-GAN) and improved dual path network (DPN) was proposed aiming at the uneven staining of pathological images and the difficulty in distinguishing between benign and malignant. Cycle-GAN was used for color normalization in order to solve the problem of low accuracy of the detection model caused by uneven staining of pathological images. The mechanism of adding small convolution, deconvolution and attention was adopted based on the DPN network by overlapping and slicing the image. The model’s ability to classify pathological image texture features was enhanced. The experimental results on the BreaKHis dataset show that the proposed algorithm effectively improves the accuracy of the classification of benign and malignant breast cancer pathological images.