Predictive modeling for cervical cancer: existing AI approaches and the emerging role of vaginal microbiome

Cervical cancer remains a major global health burden, yet current screening tools lack precision in identifying which women with high-risk human papillomavirus (HPV) infection will progress to high-grade lesions or cancer. Within a network-physiology framework, cervical carcinogenesis is viewed as emerging from dynamic interactions between viral dynamics, host immunity, vaginal ecology, vaccination status and behaviour rather than from isolated risk factors. This perspective review examines artificial intelligence (AI) approaches for cervical cancer prediction and evaluates the emerging role of the vaginal microbiome as a complementary biomarker within these interconnected physiological networks. The review synthesises evidence linking non-Lactobacillus-dominated or Lactobacillus iners-rich vaginal communities with increased HPV persistence and cervical intraepithelial neoplasia, contrasted with protective Lactobacillus crispatus-dominant communities, and outlines how these ecological signatures could be combined with HPV genotype and clinical factors in multi-modal models. A structured narrative synthesis of published AI tools demonstrates that current prognostic, diagnostic and screening algorithms rely mainly on demographic, clinical or imaging variables, with no validated models yet integrating vaginal microbiome profiles into cervical cancer risk calculators. The manuscript proposes a technical framework for microbiome-enabled modelling, covering feature engineering from community state types, algorithm selection, handling of high-dimensional omics data, and staged validation in NHS-relevant populations. Finally, it outlines a translational pathway for embedding microbiome-informed risk models into cervical screening using self-collected tampon sampling, AI-driven triage and digital decision support, and identifies key unmet needs, including longitudinal multi-omic cohorts, international consortia and robust bias auditing.