Del Hougne Michael, Del Hougne Philipp, Di Lorenzo Isabella, Höhne Christian, Schrenker Johannes, Schmitter Marc
Department of Prosthodontics, University of Würzburg, Pleicherwall 2, Würzburg, 97070, Germany.
Univ Rennes, CNRS, IETR - UMR 6164, Rennes, F-35000, France.
BMC Oral Health. 2025 Aug 31;25(1):1386. doi: 10.1186/s12903-025-06778-6.
Dental prosthesis planning is a multi-faceted and nuanced process of conceiving individual treatment plans based on dental findings and in line with established treatment guidelines. The aim of this study was to assess whether an artificial neural network (ANN) provided with sufficient training data could approximate this process.
Dental prosthesis planning was abstracted as a mapping from dental findings to choices of dental prosthesis. The problem was framed as a multi-output multi-class classification. An ANN was trained via supervised learning to approximate dental prosthesis planning based on synthetic datasets of dental findings and corresponding prosthesis choices. The accuracy on unseen test data was examined as a function of the ANN's random initializations, the training set sizes, and the ANN architecture.
Within the scope and limitations of this study, the ANN achieved an accuracy of 99.51% (± 0.15).
The ability of ANNs to learn dental prosthesis planning was confirmed within the limitations of this preliminary in-silico study. The findings of this study corroborate that ANNs have the potential to support clinicians by providing automated recommendations for choices of dental prosthesis consistent with relevant rules, ultimately supporting and enhancing clinicians' decision making. Moreover, such ANNs may, in principle, enable advanced patient self-assessment of treatment needs and improve patient care in prosthodontics.
牙修复体规划是一个多方面且细致入微的过程，需根据牙科检查结果并遵循既定治疗指南来制定个性化治疗方案。本研究的目的是评估配备足够训练数据的人工神经网络（ANN）是否能够模拟这一过程。
将牙修复体规划抽象为从牙科检查结果到牙修复体选择的映射。该问题被构建为多输出多类别分类。通过监督学习对一个人工神经网络进行训练，使其基于牙科检查结果的合成数据集及相应的修复体选择来模拟牙修复体规划。将未见过的测试数据上的准确率作为人工神经网络随机初始化、训练集大小和人工神经网络架构的函数进行检验。
在本研究的范围和局限性内，人工神经网络达到了99.51%（±0.15）的准确率。
在这项初步的计算机模拟研究的局限性内，证实了人工神经网络学习牙修复体规划的能力。本研究结果证实，人工神经网络有潜力通过根据相关规则为牙修复体选择提供自动建议来支持临床医生，最终支持并加强临床医生的决策。此外，原则上，此类人工神经网络可实现患者对治疗需求的高级自我评估，并改善口腔修复学中的患者护理。
