针对宫颈细胞病理自动筛查问题，提出一种基于深度卷积神经网络的智能辅助诊断方法。首先采用基于改进UNet深度卷积神经网络模型的语义分割方法，检测出宫颈细胞病理涂片扫描图像中的细胞(粘连簇团)区域。接着，利用VGG 16深度卷积神经网络模型，结合迁移学习技术，对检测出的细胞(粘连簇团)区域进行精确识别。为了提高深度卷积神经网络模型的性能，在进行细胞(粘连簇团)区域检测、识别的过程中，采用了数据增强技术。同时，针对该领域相关研究缺乏宫颈细胞病理液基涂片扫描图像数据集的问题，我们收集四川大学华西附二院的典型LCT筛查病例，建立了宫颈细胞病理图像HXLCT数据集，并由资深病理医生完成数据标注。实验表明，本文方法能够较好地完成宫颈细胞病理涂片扫描图像中的细胞(粘连簇团)区域检测(正确率为91.33%)，并能对检测出的区域完成正常、疑似病变二分类识别(正确率为91.6%，召回率为92.3%，ROC曲线线下面积为0.914)。本文工作将有助于宫颈细胞病理自动筛查系统的开发，对于宫颈癌早期防治具有重要意义。

Aiming to the automatic screening of cervical cytology, an intelligent auxiliary diagnosis method was proposed. At first, semantic segmentation method based on the improved UNet deep convolution neural network model was applied to detect the region of cells (cell clusters) on the scanning image of cervical cytology. Secondly, VGG 16 deep convolution neural network model combined with transfer learning was utilized for the exact diagnosis of the detected region of cells (cell clusters). The data augmentation technique was applied in the process of cells (cell clusters) detection and diagnosis to improve the performance of deep convolution neural network model. Meanwhile, considering the deficiency of scanning image datasets for the cervical cytology in the research of relevant fields, we collected the typical cases of liquid based cytology (LCT) for cervical cytological screening in West China Second Hospital of Sichuan University, set up HXLCT dataset of cervical cytological images, which were labeled by the experienced pathologists. The result of the experiment showed that, the cells (cell clusters) in the scanning images of cervical cytology could be effectively detected by the proposed method (accuracy rate was 91.6%, recall rate was 92.3%, and the area under ROC was 0.914). This method may assist in the development of automatic screening system of cervical cytology, which could be of great significance for the early prevention and treatment of endocervical carcinoma.