人工智能辅助宫颈细胞学联合人乳头瘤病毒分流用于宫颈癌筛查研究

Objective To evaluate the detection and consistency performance of artificial intelligence(AI)-assisted cervical cytology combined with human papillomavirus(HPV) triage in cervical cancer screening. Methods A total of 1,000 subjects who underwent screening at Qinghai Red Cross Hospital from January 2020 to December 2024 were included. They were divided into the AI group and the manual group,with 500 cases in each group. The two groups of subjects were independent of each other and completed the corresponding soreening procedures respectively. Both groups underwent cervical liquid-based cytology(TCT) and interpretation according to the The Bethesda system(TBS 2014), with positive cases defined as≥atypical squamous cells of undetermined significance(ASCUS). HPV typing and colposcopy/pathological verification were performed according to the protocol. Results The AI group detected 58 cases of cytological abnormalities(11.60%), significantly higher than 28 cases(5.60%) in the manual group(χ2=12.581, P<0.05). In terms of cytological grading, the AI group had 25 cases of ASCUS, 21 cases of mild squamous intraepithelial lesion(LSIL), 6 cases of atypical squamous cells with high risk of squamous intraepithelial lesion(ASC-H), 4 cases of high-grade squamous intraepithelial lesion(HSIL), and 2 cases of atypical glandular cells(AGC), the manual group had 14, 8, 2, 2 and 2 cases respectively. Pathological verification showed that the positive predictive value of the AI group for ≥HSIL was 36.21% and for ≥LSIL was 74.14%, both higher than 21.43% and 50.00% of the manual group(P<0.05). The overall positive predictive value comparison between the two groups showed 43 positive cases in the AI group and 14 positive cases in the manual group(χ2=4.923, P=0.027). The HPV positive rate in ASCUS samples of the AI group reached 78.6%, suggesting that AI interpretation has higher sensitivity in the identification of early lesions. Conclusion AI-assisted cytology combined with HPV triage can achieve higher abnormal detection and pathological consistency with the same sample size, facilitating precise stratified referral and resource allocation.