Performance of artificial intelligence-assisted cytology and high-risk human papillomavirus testing for cervical cancer screening

Evidence comparing cervical cancer screening strategies that combine artificial intelligence-assisted liquid-based cytology (AI-LBC) and high-risk human papillomavirus (HR-HPV) testing in large-scale populations is limited. This study includes 284,002 women who underwent AI-LBC and HR-HPV genotyping in community-based screening in Wuhan, China. We evaluate the performance of nine hypothesized screening strategies in detecting cervical intraepithelial neoplasia grade 2 or worse (CIN2+) and their colposcopy referral rates. After correcting for verification bias, HR-HPV testing alone (Strategy 4) demonstrates higher sensitivity (94.93% vs. 83.88%) and a higher Youden’s index (0.864 vs. 0.789) than AI-LBC alone (Strategy 1), but requires a higher referral rate (9.05% vs. 5.49%). Among the triage-based strategies, Strategy 3 (AI-LBC with reflex HR-HPV) achieves a sensitivity of 82.90%, a referral rate of 2.67%, and a Youden’s index of 0.807. HR-HPV genotyping with reflex AI-LBC (Strategy 7) demonstrates a sensitivity of 92.68%, a Youden’s index of 0.899, and a referral rate of 3.34%. Co-testing of AI-LBC and HR-HPV genotyping (Strategy 9) achieves the highest Youden’s index (0.932), with a sensitivity of 96.37% and a referral rate of 3.70%. These findings suggest that AI-LBC performs well as a standalone screening method, triage tool, or a component a co-testing strategy in cervical cancer screening. When HR-HPV genotyping is available, Strategy 7 or Strategy 9 may be considered, depending on the trade-off between detection performance and referral burden. The authors prove AI-assisted cytology effective for cervical cancer screening. When combined with HPV testing, it detects more precancerous lesions and maintains low referral rates.