EP353/#865 An automated quantitative cytology-DNA ploidy integrated analysis platform for cervical cancer screening. (4th December 2022)

• Record Type: Journal Article
• Authors: Shu, Changfa; Yu, Yang; Zhao, Meidan; Wu, Si; Gu, Pan; Zhao, Xingping; Xu, Dabao
• Is Part Of: International journal of gynecological cancer. Volume 32(2022)Supplement 3
• Page Start: A197
• Page End: A197
• Publication Date: 2022-12-04
• Subjects: Generative organs, Female -- Cancer -- Periodicals; 616.99465
• DOI: 10.1136/ijgc-2022-igcs.442
• Languages: English
• ISSNs: 1048-891X

Abstract

Abstract : Objectives: Thinprep cytology test (TCT) is a widely used method for cervical cancer screening but it is labor-intensive and lacks objectivity. Here, we aimed to establish and promote an effective TCT-based screening approach using artificial intelligence to improve the efficiency and accuracy. Methods: TCT slides were automatically scanned under microscope and images of cervical exfoliated cells were obtained. To analyze the images, artificial intelligence methods including deep convolutional neural networks were used to assist in cytology analysis and quantitative DNA ploidy analysis based on integral optical density simultaneously. Nuclear parameters such as nuclear area and perimeter were also integrated in DNA ploidy analysis to help distinguish abnormal cells. After training and validation process, the automated quantitative cytology-DNA ploidy integrated analysis (aqCDPIA) platform was established to determine the abnormity of TCT samples. The results of aqCDPIA were compared with manual TCT. Results: After examination of 21, 865 samples, aqCDPIA showed an excellent consistency of 94.6% with manual TCT results. The Kappa value was 0.733. According to the pathological results of 1, 197 samples, the sensitivities of aqCDPIA and manual TCT to discover cervical intraepithelial neoplasia were 91.4% and 88.6%, respectively. And the specificities of aqCDPIA and manual TCT were 33.4% and 41.5%. Besides, aqCDPIA has the superiority to identify non-HPV associated cervical adenocarcinoma compared with manual TCT. Conclusions: The efficient aqCDPIA platform has great potential to serve as an alternative TCT and replaces traditional visual analysis by cytopathologists. It will be beneficial to cervical cancer screening especially in the underdeveloped region where cytopathologists are scarce.