Systematic Literature Review: Machine Learning Approaches in Cervical Cancer Prediction

Abstract

Objectives: Cervical cancer continues to pose a significant global health burden, particularly in low- and middle-income countries (LMICs), where access to routine screening is limited. This systematic review aims to examine recent applications of machine learning (ML) techniques in cervical cancer prediction, with a focus on model performance, clinical applicability, and future directions. Methods: A systematic literature search was conducted across PubMed, IEEE Xplore, Scopus, and arXiv databases for studies published between January 2018 and March 2024. Inclusion criteria focused on peer-reviewed articles that applied ML methods for cervical cancer prediction and reported quantitative performance metrics. Study selection followed PRISMA guidelines, and data were extracted on ML models, datasets, evaluation metrics, and clinical relevance. Results: Out of 512 initially retrieved studies, 30 met the inclusion criteria. Convolutional Neural Networks (CNNs) showed the highest diagnostic accuracy in image-based prediction tasks, with an average Area Under the Curve (AUC) of 0.95. Ensemble learning models such as XGBoost and AdaBoost demonstrated strong performance (AUC 0.93) and offered improved interpretability. Key challenges identified include data heterogeneity, limited model explainability, regulatory hurdles, and ethical issues regarding implementation in clinical settings, particularly in LMICs. Conclusions: ML approaches, especially deep learning and ensemble methods, exhibit promising capabilities in enhancing cervical cancer prediction. However, broader clinical adoption requires addressing issues related to data diversity, transparency, regulatory compliance, and ethical deployment, with particular attention to the needs of resource-constrained environments.