Machine Vision Approaches for Cervical Cancer Screening Using Pap-Smear Images: A Systematic Review

Introduction: Cervical cancer continues to be a predominant cause of cancer-related mortality in women, especially in developing nations where its incidence is significantly raised. The Pap smear test is extensively utilized for early detection, and machine learning algorithms have evolved as effective strategies to improve diagnostic accuracy and efficacy.

Methods: This systematic review, guided by PRISMA, analyzes recent machine vision-based frameworks designed for cervical cancer screening using Pap smear images. This research was classified into three primary methodological domains: segmentation, feature extraction, and classification. Besides, commonly utilized datasets were examined to evaluate their validity and relevance for cervical cancer screening applications.

Results: This review synthesizes the methods chosen and essential elements necessary for the design of effective machine learning-based screening systems. Deep learning methods demonstrated improved accuracy when utilized with extensive annotated datasets. The absence of high-quality multi-cell datasets and inconsistencies in evaluation practices hinder comparable options among studies.

Discussion: Results show that, despite significant progress, no single framework consistently outperform others in all contexts. Deep learning-based and hybrid models demonstrate significant potential; but, variability in datasets and evaluation methods limits their generalization and clinical implementation. Improving reproducibility and reliability necessitates addressing dataset quality, annotation standards, and benchmarking practices.

Conclusion: This review considers the strengths and limitations of existing approaches and offers practical insights for the advancement of cervical cancer screening research. Future research must focus on the creation of standardized datasets and evaluation frameworks to improve clinical applicability and guarantee the robustness and scalability of diagnostic systems.