Farook Taseef Hasan, Rashid Farah, Ahmed Saif, Dudley James
PhD Scholar, Adelaide Dental School, The University of Adelaide, South Australia, Australia.
Researcher, Adelaide Dental School, The University of Adelaide, South Australia, Australia.
J Prosthet Dent. 2025 Jan;133(1):124-128. doi: 10.1016/j.prosdent.2023.01.013. Epub 2023 Feb 17.
The advent of machine learning in the complex subject of occlusal rehabilitation warrants a thorough investigation into the techniques applied for successful clinical translation of computer automation. A systematic evaluation on the topic with subsequent discussion of the clinical variables involved is lacking.
The purpose of this study was to systematically critique the digital methods and techniques used to deploy automated diagnostic tools in the clinical evaluation of altered functional and parafunctional occlusion.
Articles were screened by 2 reviewers in mid-2022 according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Eligible articles were critically appraised by using the Joanna Briggs Institute's Diagnostic Test Accuracy (JBI-DTA) protocol and Minimum Information for Clinical Artificial Intelligence Modeling (MI-CLAIM) checklist.
Sixteen articles were extracted. Variations in mandibular anatomic landmarks obtained via radiographs and photographs produced notable errors in prediction accuracy. While half of the studies adhered to robust methods of computer science, the lack of blinding to a reference standard and convenient exclusion of data in favor of accurate machine learning suggested that conventional diagnostic test methods were ineffective in regulating machine learning research in clinical occlusion. As preestablished baselines or criterion standards were lacking for model evaluation, a heavy reliance was placed on the validation provided by clinicians, often dental specialists, which was prone to subjective biases and largely governed by professional experience.
Based on the findings and because of the numerous clinical variables and inconsistencies, the current literature on dental machine learning presented nondefinitive but promising results in diagnosing functional and parafunctional occlusal parameters.
机器学习在咬合重建这一复杂领域的出现，使得有必要对用于计算机自动化成功临床转化的技术进行全面研究。目前缺乏对该主题的系统评估以及对相关临床变量的后续讨论。
本研究的目的是系统地批判在评估功能性和副功能性咬合改变的临床过程中用于部署自动化诊断工具的数字方法和技术。
2022年年中，两名审稿人根据系统评价和荟萃分析的首选报告项目（PRISMA）指南对文章进行筛选。使用乔安娜·布里格斯研究所的诊断试验准确性（JBI-DTA）方案和临床人工智能建模最低信息（MI-CLAIM）清单对符合条件的文章进行严格评估。
共提取出16篇文章。通过X线片和照片获得的下颌解剖标志点的差异在预测准确性方面产生了显著误差。虽然一半的研究采用了可靠的计算机科学方法，但缺乏对参考标准的盲法以及为了获得准确的机器学习结果而方便地排除数据，这表明传统诊断试验方法在规范临床咬合机器学习研究方面无效。由于缺乏用于模型评估的预先确定的基线或标准，很大程度上依赖于临床医生（通常是牙科专家）提供的验证，这容易产生主观偏差，并且在很大程度上受专业经验的影响。
基于研究结果，由于存在众多临床变量和不一致性，目前关于牙科机器学习的文献在诊断功能性和副功能性咬合参数方面呈现出不确定但有前景的结果。
