Related Experiment Video
Updated: Aug 2, 2025
Introduction of an Integrated Pathology Image Management, Artificial Intelligence, and Reporting System
Published on: July 11, 2025
Improving the Accuracy and Efficiency of Abnormal Cervical Squamous Cell Detection With Cytologist-in-the-Loop
Peng Xue1, Hai-Miao Xu2, Hong-Ping Tang3
1School of Population Medicine and Public Health, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China; Department of Cancer Epidemiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
Abstract:
Population-based cervical cytology screening techniques are demanding and laborious and have relatively poor diagnostic accuracy. In this study, we present a cytologist-in-the-loop artificial intelligence (CITL-AI) system to improve the accuracy and efficiency of abnormal cervical squamous cell detection in cervical cancer screening. The artificial intelligence (AI) system was developed using 8000 digitalized whole slide images, including 5713 negative and 2287 positive cases. External validation was performed using an independent, multicenter, real-world data set of 3514 women, who were screened for cervical cancer between 2021 and 2022. Each slide was assessed using the AI system, which generated risk scores. These scores were then used to optimize the triaging of true negative cases. The remaining slides were interpreted by cytologists who had varying degrees of experience and were categorized as either junior or senior specialists. Stand-alone AI had a sensitivity of 89.4% and a specificity of 66.4%. These data points were used to establish the lowest AI-based risk score (ie, 0.35) to optimize the triage configuration. A total of 1319 slides were triaged without missing any abnormal squamous cases. This also reduced the cytology workload by 37.5%. Reader analysis found CITL-AI had superior sensitivity and specificity compared with junior cytologists (81.6% vs 53.1% and 78.9% vs 66.2%, respectively; both with P < .001). For senior cytologists, CITL-AI specificity increased slightly from 89.9% to 91.5% (P = .029); however, sensitivity did not significantly increase (P = .450). Therefore, CITL-AI could reduce cytologists' workload by more than one-third while simultaneously improving diagnostic accuracy, especially compared with less experienced cytologists. This approach could improve the accuracy and efficiency of abnormal cervical squamous cell detection in cervical cancer screening programs worldwide.
