Scholarlink

<p>Cervical cancer originates in the cervix, the lower part of the uterus. It is primarily caused by persistent infection with high-risk strains of human papillomavirus. Purpose: This systematic review examines the application of deep learning methodologies for cervical cancer detection across 55 studies. The review follows the PRISMA guideline and employs the APPRAISE-AI tool to assess the methodological quality of the included papers. RQ: This study explores the main deep learning techniques applied in cervical cancer detection and analyzes how these models overcome data scarcity through augmentation strategies. It also reviews the imaging modalities employed in the literature and the extent to which explainability and interpretability are integrated into the models. Finally, the study evaluates how often healthcare professionals participate in validating these systems to support clinical decision-making. Methods: Databases of ScienceDirect, Scopus, PubMed, and Tandfoline have been utilized to find published literature between 2013 and 2023. Results: Among these, 27.27% utilized transfer learning, 56.36% employed ensemble-based approaches, and 16.36% designed custom models, demonstrating a predominant preference for ensemble methods. Addressing data scarcity, 61.81% of the studies relied on affine transformations for data augmentation. Only 1.82% used deep learning-based augmentation, highlighting a clear gap in the adoption of more advanced augmentation techniques. Publicly available datasets were utilized by 69.09% of the studies, with pap smear images being the most common modality at 65.45%. However, 70.91% of the studies did not incorporate any explainability methods, indicating a significant need for improved transparency. Additionally, only 12.73% of the studies validated their results with health practitioners, pointing to a crucial gap in clinical validation. Conclusions: These findings highlight current trends as well as persistent challenges in the field. There is a clear need for improved data augmentation strategies, wider adoption of explainability methods, and more rigorous validation practices. Together, these enhancements would significantly strengthen the reliability and clinical applicability of deep learning models for cervical cancer detection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Learning in Cervical Cancer Detection: A Systematic Review

  • 期刊论文

  • 发表日期：

    2025年12月26日

查看全文
  • Tarza Hasan Abdullah

摘要

Cervical cancer originates in the cervix, the lower part of the uterus. It is primarily caused by persistent infection with high-risk strains of human papillomavirus. Purpose: This systematic review examines the application of deep learning methodologies for cervical cancer detection across 55 studies. The review follows the PRISMA guideline and employs the APPRAISE-AI tool to assess the methodological quality of the included papers. RQ: This study explores the main deep learning techniques applied in cervical cancer detection and analyzes how these models overcome data scarcity through augmentation strategies. It also reviews the imaging modalities employed in the literature and the extent to which explainability and interpretability are integrated into the models. Finally, the study evaluates how often healthcare professionals participate in validating these systems to support clinical decision-making. Methods: Databases of ScienceDirect, Scopus, PubMed, and Tandfoline have been utilized to find published literature between 2013 and 2023. Results: Among these, 27.27% utilized transfer learning, 56.36% employed ensemble-based approaches, and 16.36% designed custom models, demonstrating a predominant preference for ensemble methods. Addressing data scarcity, 61.81% of the studies relied on affine transformations for data augmentation. Only 1.82% used deep learning-based augmentation, highlighting a clear gap in the adoption of more advanced augmentation techniques. Publicly available datasets were utilized by 69.09% of the studies, with pap smear images being the most common modality at 65.45%. However, 70.91% of the studies did not incorporate any explainability methods, indicating a significant need for improved transparency. Additionally, only 12.73% of the studies validated their results with health practitioners, pointing to a crucial gap in clinical validation. Conclusions: These findings highlight current trends as well as persistent challenges in the field. There is a clear need for improved data augmentation strategies, wider adoption of explainability methods, and more rigorous validation practices. Together, these enhancements would significantly strengthen the reliability and clinical applicability of deep learning models for cervical cancer detection.