Toffaha Khaled, Simsekler Mecit Can Emre, Sleptchenko Andrei, Kortt Michael A, Bukasa Laurette L
Department of Management Science & Engineering, Khalifa University of Science & Technology, Abu Dhabi, United Arab Emirates.
Abu Dhabi Health Data Services, M42, Abu Dhabi, United Arab Emirates.
J Multidiscip Healthc. 2025 Aug 25;18:5199-5211. doi: 10.2147/JMDH.S524132. eCollection 2025.
Cervical cancer remains a major global health challenge, necessitating enhanced risk stratification and early detection methodologies. This study proposes a comprehensive predictive framework for cervical cancer leveraging advanced machine learning (ML) algorithms and Bayesian Belief Networks (BBNs), illustrating the transformative role of digital technologies in healthcare and education within an increasingly digitized society.
A cohort of 858 patients was analyzed, addressing data challenges, including missing values, class imbalance, and nonlinear feature interactions, that frequently compromise the reliability of predictive modeling. Methodologically, this study integrated advanced data science approaches, including multiple imputation, feature selection, and imbalance mitigation, advancing medical analytics to ensure robust model generalizability.
High predictive performance was observed across different cervical cancer screening tests. The combined target ML model achieved an accuracy of 95.6%, an area under the receiver-operating characteristic curve (AUROC) of 0.958, and an F1-score of 0.945. The BBN, built upon the Bayesian Additive Regression Trees (BART) model, demonstrated a positive prediction rate (sensitivity) of 91.3% and a negative prediction rate (specificity) of 86.8%.
These results validate the technical efficacy of the proposed framework and underscore its potential for integration into clinical decision-support systems. Beyond clinical applications, this research contributes to computational oncology by demonstrating the synergistic potential of combining probabilistic graphical models with ML techniques. The study highlights the critical role of interdisciplinary collaboration between clinical experts and data scientists in creating effective AI healthcare solutions. It also emphasizes the need for upskilling healthcare workers and optimizing healthcare delivery processes to fully realize the benefits of precision medicine.
宫颈癌仍然是一项重大的全球健康挑战，因此需要加强风险分层和早期检测方法。本研究提出了一个利用先进机器学习（ML）算法和贝叶斯信念网络（BBN）的宫颈癌综合预测框架，说明了数字技术在日益数字化的社会中对医疗保健和教育的变革性作用。
分析了一组858名患者的数据，解决了包括缺失值、类别不平衡和非线性特征交互等经常影响预测建模可靠性的数据挑战。在方法上，本研究整合了先进的数据科学方法，包括多重插补、特征选择和不平衡缓解，推进医学分析以确保强大的模型泛化能力。
在不同的宫颈癌筛查测试中观察到了较高的预测性能。组合目标ML模型的准确率达到95.6%，受试者工作特征曲线下面积（AUROC）为0.958，F1分数为0.945。基于贝叶斯加法回归树（BART）模型构建的BBN显示出91.3%的阳性预测率（敏感性）和86.8%的阴性预测率（特异性）。
这些结果验证了所提出框架的技术有效性，并强调了其整合到临床决策支持系统中的潜力。除了临床应用，本研究通过展示概率图形模型与ML技术相结合的协同潜力，为计算肿瘤学做出了贡献。该研究强调了临床专家和数据科学家之间跨学科合作在创建有效的人工智能医疗保健解决方案中的关键作用。它还强调了提高医护人员技能和优化医疗保健交付流程以充分实现精准医学益处的必要性。
