Application of Artificial Intelligence in HPV Related Cervical Cancer Diagnosis: A Systematic Review

Objectives: This review aims to quantify AI diagnostic performance specifically for HPV-related cervical cancer detection, stratified by methodological category, and to evaluate the added value of HPV genotype integration. Methods: A systematic review was conducted following PRISMA 2020 guidelines. SciSpace, PubMed/MEDLINE, Google Scholar, and ArXiv were searched for records published between January 2015 and May 2025, yielding 647 records; after deduplication (n=450) and two-stage screening, 37 studies met inclusion criteria. Two reviewers independently extracted study design, AI methodology, dataset characteristics, and performance metrics; quality was appraised using QUADAS-2 and PROBAST. Findings: Deep learning dominated the included literature (24/37, 64.9%), followed by classical machine learning (8/37, 21.6%) and multimodal fusion (5/37, 13.5%). Reported accuracy spanned 83.0-99.99%, sensitivity 80.0-100%, specificity 67.0-99.0%, and AUC 0.85-0.96. Multimodal models fusing image data with HPV genotype information achieved the highest ceiling performance (AUC up to 0.963), exceeding image-only deep learning and expert-clinician benchmarks reported within the same studies. Novelty: Unlike prior broad reviews of AI in cervical cancer, this is the first systematic review to isolate HPV genotype-linked diagnostic AI as a distinct analytic unit, the first in this space to apply dual formal risk-of-bias appraisal (QUADAS-2, PROBAST) across all included studies, and the first to quantify the performance differential between genotype-fused multimodal models and image-only approaches (AUC up to 0.963 versus 0.96). This stratification provides a quantitative framework for evaluating the added value of multimodal architectures, synthesised into a translational roadmap for future AI development.