Jennifer Bailón Mieles, Jessica Pilozo
Colposcopy has limitations due to observer variability, which is why artificial intelligence (AI) has been proposed to improve its accuracy. This systematic review (PRISMA 2020) analyzed the diagnostic accuracy of AI systems on colposcopic images to detect high-grade cervical intraepithelial lesions (CIN2+, CIN3+, HSIL). Searches were conducted in PubMed, Scopus, Web of Science, ScienceDirect, IEEE Xplore, and Google Scholar (January 2021 – June 2026), including studies with histopathological confirmation as the reference standard and assessing quality with QUADAS-2. Ten studies were included (over 15,000 images and 7,500 patients), mostly retrospective and from Asia. Performance was heterogeneous: sensitivity (63.0–99.9%), specificity (57.6–99.8%), and AUC up to 0.95. AI proved useful as a supportive tool, especially for junior colposcopists. The main biases were related to patient selection and retrospective design. In conclusion, AI shows promising diagnostic accuracy for high-grade lesions, although with significant variability. Prospective, multicenter studies conducted in real-world settings are needed to confirm its effectiveness and cost-effectiveness before widespread clinical implementation.