Bringing Cervical Cancer Screening Closer to Women: Feasibility of Artificial Intelligence and Remote Assessment in Primary Health Care

Abstract

Objective: The objective was to assess the feasibility of image-based methods for screening and triaging women in a single visit by: (i) a trained but inexperienced nurse, (ii) remote expert review via a web system, (iii) an artificial intelligence (AI) model. Methods: Sexually active, non-pregnant women (25–65 years) were screened using visual inspection method Cervical images captured with Smart Scope® CX were assessed independently by nurses, remote experts, and AI, with assessors blinded to each other. Referrals for colposcopy were based on remote expert evaluations followed by colposcopy/biopsy. Results: Among 871 women screened, AI identified 205 positives; experts identified 201. Colposcopy was performed on 69 women, 40 of them had a biopsy. Compared to histopathology, AI achieved 86.7% sensitivity, 92.0% specificity, and 90.0% accuracy (AUC = 0.894). Remote experts showed high sensitivity (86.7%) but low specificity (32%) and accuracy (52.5%). Conclusion: This study provides proof of concept for the feasibility of the AI-driven Smart Scope® CX test as a single-visit “screen-and-triage” tool in primary healthcare settings. Additionally, remote expert assessment demonstrating performance comparable to colposcopy indicates its potential as an alternative triage method in low-resource settings.