Related Experiment Video
Updated: May 10, 2026
Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model
Published on: April 18, 2025
Multi-center-validated machine learning model for cervical cancer based on human papillomavirus genotyping results
Ziren Chen1, Yihui Zeng1, Lingxiao Luo1
1Obstetrics and Gynecology Hospital of Fudan University, Department of Cervical Disease Center, Shanghai, China.
Objective:
Cervical cancer remains a leading cause of gynecologic cancer-related mortality worldwide, with human papillomavirus (HPV) testing being a key screening tool. However, current guidelines often categorize high-risk HPV types broadly, overlooking their varying oncogenic potentials. This study aimed to develop and validate a machine learning model to predict cervical intra-epithelial neoplasia grade (CIN) 2 or worse (CIN2+) using HPV genotyping, cytology results, and patient age.
Methods:
A retrospective analysis was conducted using data from 52,063 patients (61,022 clinical records) from the Obstetrics and Gynecology Hospital of Fudan University, Shanghai, collected between October 2017 and June 2023. Clinical data included HPV genotyping, cytology results, and histopathologic diagnoses. Six machine learning algorithms (logistic regression, decision tree, random forest, Support Vector Machine-Radial Basis Function, XGBoost, and CatBoost) were compared, with performance evaluated using area under the curve, accuracy, sensitivity, and F1 score. The optimal model was validated using a temporal cohort from Obstetrics and Gynecology Hospital of Fudan University and external cohorts from Xiamen Maternal and Child Health Hospital and Jinan Maternal and Child Health Hospital.
Results:
Among the models, CatBoost achieved the highest area under the curve (0.917) in internal validation, with an accuracy of 86% and sensitivity of 78% for CIN2+. External validation confirmed robust performance: 89% accuracy (Obstetrics and Gynecology Hospital of Fudan University), and 78% (Xiamen Maternal and Child Health Hospital), and 80% (Jinan Maternal and Child Health Hospital). Simulation analysis indicated that the AI (Artificial Intelligence)-guided strategy could reduce colposcopy referrals by 14.3%, whereas maintaining 96.8% sensitivity and an exceptional negative predictive value of 99.18%. A user-friendly tool (www.cervixcare.cn) was developed to provide individualized CIN2+ risk scores based on input variables.
Conclusions:
This multi-center-validated model enables precise risk stratification by accounting for HPV genotype-specific pathogenicity, improving clinical decision-making for cervical cancer screening and management.
