基于生成对抗网络的低分化宫颈癌病理图像分类

使用生成对抗网络(GAN)扩充宫颈癌病理图像的数据集以提高计算机辅助诊断的准确率.首先，使用GAN进行细胞质部分图像生成；其次，使用两次k-means聚类对生成图像进行筛选；最后，使用Inception-V3模型对数据集进行分类训练.结果表明，在测试集相同的情况下，该方法可以将总体分类准确率提升约2.5%，尤其对低分化宫颈癌病理图像有显著效果.通过GAN解决了组织病理学图像无方向性、内容复杂、前景目标规则性差等问题，证明了该方法的有效性及发展潜力.

Generative Adversarial Networks Based Pathological Images Classification of Poorly Differentiated Cervical Cancer

The accuracy of computer-assisted diagnosis can be improved by using generative adversarial networks(GAN) to extend the data set of cervical cancer patholigical images. First， the cytoplasmic part of the histopathological images was generated by GAN; then， k-means clustering was used twice to select images generated by GAN; finally， Inception-V3 model was used to train a classifier. The results showed that the accuracy is improved by an average of 2.5% under the same test data set. Especially， it has significant effect for poorly differentiated cervical cancer pathological images. The non-directionality， complexity of content and poor regularity of foreground target for histopathological images are solved by GAN， which proves the effectiveness and the potential of this method.