人工智能阴道镜在大规模宫颈癌筛查异常转诊人群中的价值

The value of artificial intelligence colposcopy in abnormal referral populations during large-scale cervical cancer screening

摘要

目的 评估人工智能阴道镜在宫颈癌筛查异常转诊中的诊断价值。方法 基于深度学习算法开发人工智能阴道镜系统,选择2021年7月至2024年2月于本院收治的400例自愿参加免费宫颈癌筛查的女性为研究对象并随机分为两组,200例进行人工阴道镜活检纳入人工组,200例使用人工智能(AI)辅助诊断软件进行AI阴道镜活检纳入AI组.以病理结果为金标准进行诊断比较.通过受试者工作特征(ROC)曲线分析诊断效能,采用Kappa指数进行一致性检验。结果 两组基线特征未见显著统计学差异.AI组对HSIL+的诊断符合率显著高于人工组(89.00% vs. 82.00%,χ²=3.95,P=0.047),两组在其他病理分级的诊断符合率及总符合率比较,差异均无统计学意义(P>0.05).ROC曲线分析结果显示:人工阴道镜诊断宫颈癌的曲线下面积(AUC)值优于AI组[AUC及95%CI分别为0.942(0.833~1.000)、0.786(0.640~0.932)];人工阴道镜诊断HSIL的AUC值仍优于AI组[AUC及95%CI分别为0.850(0.793~0.906)、0.741(0.670~0.813)].AI组展现更高特异性(84.30% vs. 76.80%,χ²=4.27,P=0.039)和更低假阳性率(15.70% vs. 23.20%,χ²=5.84,P=0.016).LSIL及炎症诊断中AI组的灵敏度(52.10% vs. 73.80%,χ²=12.45,P<0.001)与特异度(68.40% vs. 82.30%,χ²=9.67,P=0.002)均显著低于人工组.人工组Kappa值更高(0.67 vs. 0.53),即人工组诊断一致性上优于AI,AI组在复杂病例中表现不足。结论 AI组在宫颈癌和HSIL诊断中符合率和特异性显著优于人工组,AI阴道镜在大规模数据中表现出较高特异性和较低假阳性率.随着技术优化,AI有望提升低级别病变识别和诊断一致性,二者优势互补可为异常转诊患者提供更精准的诊断支持.

Abstract

Objective To evaluate the diagnostic value of artificial intelligence (AI) colposcopy in abnormal referrals during cervical cancer screening. Methods An AI colposcopy system was developed based on deep learning algorithms. A total of 400 women who voluntarily participated in free cervical cancer screening at our hospital between July 2021 and February 2024 were enrolled and randomly divided into two groups. The conventional colposcopy biopsy group (n=200) was designated as the manual group, and the AI-assisted diagnostic software colposcopy biopsy group (n=200) was designated as the AI-assisted group. Pathological results were used as the gold standard for diagnostic comparison. The diagnostic performance was evaluated using receiver operating characteristic (ROC) curve analysis, and consistency between methods was assessed using the Kappa statistic. Results No significant differences were observed in baseline characteristics between the two groups. The diagnostic concordance rate for high-grade squamous intraepithelial lesion (HSIL)+ in the AI-assisted group was significantly higher than that in the manual group (89.00% vs. 82.00%, χ²=3.95, P=0.047). For other pathological grades and overall diagnostic concordance, no statistically significant differences were found between the two groups (P>0.05). ROC curve analysis showed that the area under the curve (AUC) for cervical cancer diagnosis using manual colposcopy was superior to that of the AI-assisted group [AUC (95% CI): 0.942 (0.833-1.000) and 0.786 (0.640-0.932), respectively]. Similarly, the AUC for HSIL diagnosis in the manual group remained higher than that of the AI-assisted group [AUC (95% CI): 0.850 (0.793-0.906) and 0.741 (0.670-0.813), respectively]. The AI-assisted group displayed higher specificity (84.30% vs. 76.80%, χ²=4.27, P=0.039) and a lower false-positive rate (15.70% vs. 23.20%, χ²=5.84, P=0.016). However, for low-grade squamous intraepithelial lesions (LSIL) and inflammatory lesions, both the sensitivity (52.10% vs. 73.80%, χ²=12.45, P<0.001) and specificity (68.40% vs. 82.30%, χ²=9.67, P=0.002) of the AI-assisted group were significantly lower than those of the manual group. The manual group achieved a higher weighted Kappa coefficient (0.67 vs. 0.53), indicating superior diagnostic consistency over AI, which underperformed in complex cases requiring nuanced pathological interpretation. Conclusion The AI-assisted group demonstrated significantly higher concordance and specificity in the diagnosis of cervical cancer and HSIL compared to the human group. AI-assisted colposcopy showed high specificity and a lower false-positive rate in large-scale data. With further optimization of the technology, AI has the potential to improve the identification of low-grade lesions and improve diagnostic consistency. The complementary advantages of both methods can provide more accurate diagnostic support for patients referred due to abnormalities.

关键词

人工智能 / 阴道镜 / 宫颈癌筛查 / 转诊 / 诊断价值

Key words

artificial intelligence / colposcopy / cervical cancer screening / referral / diagnostic value

参考文献

[1] 陈飞, 尤志学, 隋龙, 等. 阴道镜应用的中国专家共识[J]. 中华妇产科杂志, 2020, 55(7): 443-449.

[2] Green L I, Mathews C S, Waller J, et al. Attendance at early recall and colposcopy in routine cervical screening with human papillomavirus testing[J]. Int J Cancer, 2021, 148(8): 1850-1857.

[3] Roopa H, Mittal S, Basu P. Role of colposcopy in the management of women with abnormal cytology[J]. Cytojournal, 2022, 19: 40.

[4] Xue P, Ng M T A, Qiao Y. The challenges of colposcopy for cervical cancer screening in LMICs and solutions by artificial intelligence[J]. BMC Med, 2022, 18(1): 169.

[5] 薛鹏. 基于人工智能的电子阴道镜辅助诊断系统构建及多中心应用评估[D]. 中国医学科学院北京协和医学院, 2020.

[6] Hu L, Bell D, Antani S, et al. An observational study of deep learning and automated evaluation of cervical images for cancer screening[J]. J Natl Cancer Inst, 2019, 111(9): 923-932.

[7] 陈明扬, 蔡紫庭, 薛鹏, 等. 人工智能在肿瘤研究和临床中的应用[J]. 基础医学与临床, 2022, 42(11): 1637-1643.

[8] 薛鹏. 人工智能技术在宫颈癌筛查、分流和诊断中的应用研究[D]. 北京: 中国医学科学院, 2023.

[9] Bray F, Ferlay J, Soerjomataram I, et al. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries[J]. CA Cancer J Clin, 2018, 68(6): 394-424.

[10] 臧春宝, 吴爱林, 王慧妍, 等. 磷酸铝凝胶联合康复新液保留灌肠预防宫颈癌放疗后放射性肠炎的临床研究[J]. 蚌埠医学院学报, 2019, 44(8): 1004-1008.

[11] 薛鹏, 唐朝, 乔友林, 等. 人工智能电子阴道镜辅助诊断系统对宫颈癌筛查的现实挑战和未来机遇[J]. 中国肿瘤, 2019, 28(7): 483-486.

[12] Wu A, Xue P, Abulizi G, et al. Artificial intelligence in colposcopic examination: a promising tool to assist junior colposcopists[J]. Front Med (Lausanne), 2023, 10: 1060451.

[13] 陈苗, 高丽, 张婷, 等. 1357例宫颈上皮内瘤变临床特征及阴道镜图像的诊断意义[J]. 中国妇幼健康研究, 2023, 34(2): 59-63.

[14] 薛鹏, 乔友林, 江宇. 人工智能在医学内窥镜诊断中的应用[J]. 中华肿瘤杂志, 2018, 40(12): 890-893.

[15] 何雪梅, 黄辉, 王婷, 等. 人工智能辅助宫颈细胞学筛查联合HPV分流在宫颈癌人群筛查中的应用分析[J]. 徐州医科大学学报, 2022, 42(4): 273-278.

[16] Hussain E, Mahanta L B, Das C R, et al. A comprehensive study on the multi-class cervical cancer diagnostic prediction on pap smear images using a fusion-based decision from ensemble deep convolutional neural network[J]. Tissue Cell, 2020, 65: 101347.

[17] 车拴龙, 刘栋, 刘斯, 等. 人工智能技术在宫颈细胞筛查中的应用进展和挑战[J]. 中华临床实验室管理电子杂志, 2019, 7(4): 193-198.

[18] 李艳辉, 周星, 徐琴, 等. 基于云技术和人工智能的自动化细胞学诊断平台在703103名女性宫颈癌筛查中的应用研究[J]. 中国妇产科临床杂志, 2022, 23(4): 364-368.

[19] 中国妇幼保健协会, 中国妇幼保健协会妇女病防治专业委员会. 人工智能辅助宫颈细胞学诊断技术的应用及质量控制专家共识[J]. 中国妇幼健康研究, 2023, 34(12): 7-13.

基金资助

2021年度中国妇幼保健协会科研创新项目(ZGFYBJXHKY2102)