Assessing artificial intelligence enabled liquid-based cytology for triaging HPV-positive women: a population-based cross-sectional study

Introduction: Cytology-based triaging is commonly used to manage the care of women with positive human papillomavirus (HPV) results, but it suffers from subjectivity and a lack of sensitivity and reproducibility. The diagnostic performance of an artificial intelligence-enabled liquid-based cytology (AI-LBC) triage approach remains unclear. Here, we compared the clinical performance of AI-LBC, human cytologists and HPV16/18 genotyping at triaging HPV-positive women. Material and methods: HPV-positive women were triaged using AI-LBC, human cytologists and HPV16/18 genotyping. Histologically confirmed cervical intraepithelial neoplasia grade 2/3 or higher (CIN2+/CIN3+) were accepted as thresholds for clinical performance assessments. Results: Of the 3514 women included, 13.9% (n = 489) were HPV-positive. The sensitivity of AI-LBC was comparable to that of cytologists (86.49% vs 83.78%, P = 0.744) but substantially higher than HPV16/18 typing at detecting CIN2+ (86.49% vs 54.05%, P = 0.002). While the specificity of AI-LBC was significantly lower than HPV16/18 typing (51.33% vs 87.17%, P < 0.001), it was significantly higher than cytologists at detecting CIN2+ (51.33% vs 40.93%, P < 0.001). AI-LBC reduced referrals to colposcopy by approximately 10%, compared with cytologists (51.53% vs 60.94%, P = 0.003). Similar patterns were also observed for CIN3+. Conclusions: AI-LBC has equivalent sensitivity and higher specificity compared with cytologists, with more efficient colposcopy referrals for HPV-positive women. AI-LBC could be particularly useful in regions where experienced cytologists are few in number. Further investigations are needed to determine triaging performance through prospective designs.