基于图神经网络改进宫颈癌筛查系统与宫颈细胞DNA定量在高危型HPV阳性患者中的比较

Cervical cancer screening system TruScreen with graph neural network-based classification versus HPV DNA quantification in high-risk HPV-positive patients

摘要:

目的 基于图神经网络对比分析宫颈癌筛查系统（truScreen，TS）、宫颈细胞DNA定量检查高危型人乳头瘤病毒（human papillomavirus，HPV）阳性患者中筛查宫颈癌及癌前病变的效果。方法 选取2022年1月至2023年5月于重庆大学附属黔江医院妇科门诊行宫颈癌高危型HPV检测阳性的400例患者为研究对象。对所有研究对象依次进行TS、宫颈细胞DNA定量、阴道镜下组织病理学检查。以组织病理检查为“金标准”，除采用TS系统已有模块外，基于图神经网络进行标识学习，提取特征，随后使用支持向量机（support vector machines，SVM）和随机森林（random forest，RF）等分类器进行分类，计算并比较 TS、宫颈细胞DNA定量筛查宫颈癌及癌前病变的灵敏度、特异度、受试者工作特征（receiver operating characteristic，ROC）曲线下面积（area under the curve，AUC）、阴性预测值、阳性预测值、与活检病理结果一致性检验统计量Kappa值。结果 TS融合图神经网络筛查宫颈癌及癌前病变的灵敏度、特异度、AUC、阴性预测值、阳性预测值分别为93.75%、95.45%、0.97、98.18%、87.50%，均高于宫颈细胞DNA定量检查（分别为81.25%、90.91%、0.92、96.15%、64.29%）。与活检病理结果的一致性检验显示，TS融合图神经网络与活检病理结果的Kappa值为0.89，高于宫颈细胞DNA定量检查（0.72）。ROC曲线下面积的比较显示，TS融合图神经网络与宫颈细胞DNA定量检查之间存在显著差异。结论 TS融合图神经网络在高危型HPV阳性患者中筛查宫颈癌及癌前病变的效果优于宫颈细胞DNA定量检查，具有较高的准确性和可靠性，可作为一种简单、有效的宫颈癌筛查方法。

Abstract:

Objective To compare the effectiveness of TruScreen（TS） integrated with a graph neural network and human papillomavirus（HPV） DNA quantification in cervical precancer and cancer screening in high-risk HPV-positive patients.Methods We included 400 patients whotested positive for high-risk HPV subtypes at the Department of Gynecology，Qianjiang Hospital，Chongqing University from January 2022 to May 2023. All the patients underwent TS，HPV DNA quantification in cervical cells，and a colposcopy-directed biopsy. The biopsy pathology results were used as the gold standard. The TS system was combined with graph neural network-based representation learning and feature extraction and classification with support vector machine and random forest classifiers. We calculated and compared the sensitivity，specificity，area under the receiver operating characteristic curve（AUC），negative predictive value，positive predictive value，agreement with biopsy pathology （kappa value） of the modified TS system and HPV DNA quantification in cervical precancer and cancer screening.Results The sensitivity，specificity，AUC，negative predictive value，and positive predictive value of TS with graph neural network-based classification for cervical precancer and cancer screening were 93.75%，95.45%，0.97，98.18%，and 87.50%，respectively，which were higher than those of HPV DNA quantitative testing （81.25%，90.91%，0.92，96.15%，and 64.29%，respectively）. The kappa coefficient of agreement of TS with graph neural network-based classification with biopsy pathology was 0.89，higher than 0.72 of HPV DNA quantitative testing with biopsy pathology. There was a significant difference in the AUC of TS with graph neural network-based classification and HPV DNA quantitative testing （P<0.05）.Conclusion TS with graph neural network-based classification has better performance in screening for cervical precancer and cancer in high-risk HPV-positive patients than HPV DNA quantitative testing，with high accuracy and reliability. It can be used as a simple and effective cervical cancer screening method.