Agreement Between the CerviCARE AI Artificial-Intelligence Software and Visual Inspection with Acetic Acid (VIA) for Cervical Cancer Screening in Women Aged Over 18 Years in Vinh Long Province, Vietnam: A Community-Based Cross-Sectional Study

Résumé fourni par la source

Background: Cervical cancer is preventable, yet screening coverage remains low in low- and middle-income countries (LMICs), where HPV DNA testing and cytology are often unavailable and visual inspection with acetic acid (VIA) is the most widely used method. Artificial-intelligence (AI) image-analysis tools may support screening, so their agreement with VIA is of practical relevance. Objective: To describe cervical cancer screening results with CerviCARE AI, VIA, and Pap smear, and to evaluate the agreement between CerviCARE AI and VIA in women aged ≥18 years in Vinh Long Province, Vietnam, in 2025. Materials and Methods: A community-based cross-sectional study included 2757 women, of whom 2316 underwent all three methods. Agreement was assessed with Cohen's kappa (asymptotic and bootstrap 95% CIs), the exact McNemar test, and the prevalence-adjusted bias-adjusted kappa (PABAK). Results: Among 2316 women, CerviCARE AI classified 360 (15.5%) as abnormal, VIA was positive in 413 (17.8%), and and Pap smear was positive in only 4 (0.2%). CerviCARE AI and VIA showed 97.71% observed agreement and a Cohen's kappa of 0.918 (95% CI: 0.896-0.940; bootstrap 0.896-0.938), indicating almost perfect agreement, with a PABAK of 0.954. All 360 AI-abnormal cases were VIA-positive, whereas VIA flagged an additional 53 cases (exact McNemar p < 0.001), so AI-positive results were a subset of VIA. Agreement remained almost perfect across age strata (κ 0.83-1.00). In the overall questionnaire sample, only 2.6% of women had ever received HPV vaccination and 32.6% had ever been screened, although 98.0% considered screening necessary and 91.7% were willing to be vaccinated; cost was the most frequent barrier (49.0%). Conclusion: CerviCARE AI showed almost perfect agreement with VIA under community screening conditions and classified fewer cases as abnormal than VIA. These findings indicate feasibility and concordance rather than diagnostic accuracy and require validation against histopathology or HPV DNA testing before conclusions on diagnostic performance can be drawn.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.