Recent Advances in AI-Based Cervical Cancer Prediction: A Comprehensive Systematic Review

Abstract

Cervical cancer is still a major cause of cancer-related morbidity and mortality in women globally, and especially in low- and middle-income countries where access to organized screening and early diagnostic services is limited. Artificial Intelligence (AI) has advanced cervical cancer prediction by improving automated identification, risk assessment, image analysis, and clinical decision-making. This systematic review integrates findings from 80 studies published over the last five years (2021 to 2026), following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The most relevant studies were obtained from major scientific databases: Scopus, IEEE Xplore, SpringerLink, ScienceDirect, PubMed, ACM Digital Library, MDPI, and Google Scholar. The literature was then thoroughly analyzed based on the AI methodologies, datasets used, prediction performance, explainability approaches, and how the predictions were applied in cervical cancer. The review then methodically analyzed the selected literature based on the AI methodologies/approaches used, the datasets used, the predictive performance of the AI models, the explainability approaches used, and how explainability was applied in cervical cancer.

The review also shows a marked shift from traditional machine learning algorithms, including Random Forest, Support Vector Machine, and XGBoost, to cutting-edge deep learning structures, transfer learning, ensemble learning, transformer-based models, and Explainable Artificial Intelligence (XAI). The UCI Cervical Cancer Risk Factors and Herlev or SIPaKMeD datasets are the most popular public resources, and multimodal data integration is a new research area. Most deep learning and ensemble models reported an accuracy >95%, achieving considerable improvements in diagnostic performance. However, several issues, such as the restricted diversity of the datasets, class imbalance, lack of external validation, computational complexity, and lack of clinical deployment, remain major challenges for actual implementation. The results suggest that future studies focus on multimodal, explainable, privacy-preserving, and clinically validated AI frameworks supported by large multicenter datasets and prospective clinical evaluation. These advances can lead to better early detection, individualized treatment planning, and worldwide work towards cervical cancer eradication.