INTRODUCTION: Cervical cancer disproportionately affects women in low- and middle-income countries (LMICs), where access to cytology, human papillomavirus (HPV) DNA testing, and expert colposcopy remains limited. Advances in smartphone imaging and artificial intelligence (AI) algorithms offer a scalable, low-cost alternative for screening. We present the protocol for a systematic review and meta-analysis evaluating the diagnostic accuracy of smartphone-based AI colposcopy in LMICs, with preliminary findings from our ongoing review. METHODS: This review is registered in PROSPERO (ID CRD420251142715) and follows PRISMA-DTA guidelines. Eligible studies include randomized and non-randomized diagnostic accuracy studies assessing smartphone-based AI colposcopy or visual inspection with acetic acid (VIA) compared with reference standards (histopathology, expert colposcopy, cytology, HPV DNA testing). Databases searched are PubMed/MEDLINE, Embase, CENTRAL, CINAHL, LILACS, Scopus, and IEEE Xplore (2010 to present). Dual independent reviewers conduct screening, extraction, and bias assessment using QUADAS-2. Pooled sensitivity and specificity will be estimated using a bivariate random-effects model, with summary ROC curves generated. RESULTS: As of January 2025, more than 1,200 records were identified, with 46 studies undergoing full-text review. Early synthesis suggests smartphone-based AI achieves high diagnostic accuracy, with sensitivities of 85–98% and specificities of 75–92% versus histopathology or expert colposcopy. Feasibility studies emphasize offline functionality, task-shifting potential to nonspecialists, and strong patient acceptability. CONCLUSIONS/IMPLICATIONS: Smartphone-based AI colposcopy shows strong promise for expanding cervical cancer screening in resource-limited settings. This review will provide pooled diagnostic accuracy estimates and identify research gaps, informing clinical implementation and policy in line with the World Health Organization’s elimination strategy.
Nicole Friedlich (Thu,) studied this question.