Background: Low- and middle-income countries (LMICs) bear a disproportionate burden of cervical cancer, compounded by limited access to organised screening, diagnostic confirmation, referral, treatment, and follow-up. The World Health Organization (WHO) recommends high-risk human papillomavirus (HPV) DNA testing as the preferred primary screening approach where feasible, while visual inspection with acetic acid (VIA) remains a context-dependent alternative. Automated visual evaluation (AVE), an artificial intelligence (AI)-assisted image-based approach, offers potential improvements in screening reproducibility and risk stratification. However, diagnostic accuracy alone does not establish clinical readiness or population-level benefit.
Objectives: To evaluate the current evidence supporting AVE within HPV-based cervical screening pathways in LMICs and identify the validation, clinical workflow, safety, and health-system requirements necessary for responsible implementation.
Materials and Methods: A focused narrative synthesis, rather than a systematic review, was conducted using WHO cervical screening guidance and relevant publications identified through PubMed/MEDLINE, PubMed Central, and WHO online repositories. Evidence concerning AVE, HPV DNA testing, VIA, image quality, and implementation in resource-constrained settings was examined. Studies were evaluated according to population characteristics, HIV/HPV status, imaging devices, reference standards, validation methods, diagnostic thresholds, and potential biases, including selective histopathological verification. Findings were synthesised across diagnostic validity, clinical workflow, patient safety, workforce requirements, infrastructure, affordability, equity, referral-to-treatment pathways, and governance.
Results: Current evidence supports AVE primarily as a clinician-assisted triage and risk-stratification tool, particularly among HPV-positive women, rather than as a standalone diagnostic method. In a prospective public-sector study involving 8,204 women in Zambia, approximately 54% of whom were living with HIV, AVE achieved an internal-validation area under the receiver operating characteristic curve (AUC) of 0.91 for detecting cervical intraepithelial neoplasia grade 2 or worse (CIN2+), with 85% sensitivity and 86% specificity using Samsung Galaxy A21s images. Among HPV-positive women, the triage AUC was 0.87. However, performance declined to an AUC of 0.79 using J8 images, highlighting device-dependent variability and concerns regarding generalisability. Selective histopathological verification further limited interpretation.
The PAVE study demonstrated that combining HPV genotype groups with AVE categories generated 12 risk strata for CIN3+, ranging from 0% to 29%, compared with 1.1%–13.8% for AVE alone and 3.6%–20.8% for HPV genotyping alone. Although these findings suggest improved risk stratification, the combined strategy was not evaluated for real-time clinical management. Evidence remains insufficient regarding external validity, treatment completion, affordability, cost-effectiveness, equity, and patient-important outcomes.
Conclusion: AVE represents a promising adjunct to HPV-centred cervical screening in LMICs, particularly for clinician-supported triage and risk stratification. Nevertheless, promising diagnostic performance does not establish clinical readiness. Responsible implementation requires prospective, locally led studies evaluating external and device-specific validation, image-quality assessment, deferral safeguards, clinician oversight, workflow integration, referral and treatment completion, affordability, equity, and post-deployment monitoring. AVE should complement rather than replace established diagnostic and treatment pathways, with adoption guided by demonstrated clinical effectiveness, patient safety, and health-system readiness.