Related Experiment Video
Updated: Jul 4, 2025
A Postoperative Evaluation Guideline for Computer-Assisted Reconstruction of the Mandible
Published on: January 28, 2020
Convolutional neural network-based measurement of crown-implant ratio for implant-supported prostheses
Jin-Ping Zhang1, Ze-Hui Wang2, Juan Zhang3
1Postgraduate student, Department of Oral Implantology, Affiliated Hospital of Stomatology, Nanjing Medical University, Nanjing, PR China.
Statement Of Problem:
Research has revealed that the crown-implant ratio (CIR) is a critical variable influencing the long-term stability of implant-supported prostheses in the oral cavity. Nevertheless, inefficient manual measurement and varied measurement methods have caused significant inconvenience in both clinical and scientific work.
Purpose:
This study aimed to develop an automated system for detecting the CIR of implant-supported prostheses from radiographs, with the objective of enhancing the efficiency of radiograph interpretation for dentists.
Material And Methods:
The method for measuring the CIR of implant-supported prostheses was based on convolutional neural networks (CNNs) and was designed to recognize implant-supported prostheses and identify key points around it. The experiment used the You Only Look Once version 4 (Yolov4) to locate the implant-supported prosthesis using a rectangular frame. Subsequently, two CNNs were used to identify key points. The first CNN determined the general position of the feature points, while the second CNN finetuned the output of the first network to precisely locate the key points. The network underwent testing on a self-built dataset, and the anatomic CIR and clinical CIR were obtained simultaneously through the vertical distance method. Key point accuracy was validated through Normalized Error (NE) values, and a set of data was selected to compare machine and manual measurement results. For statistical analysis, the paired t test was applied (α=.05).
Results:
A dataset comprising 1106 images was constructed. The integration of multiple networks demonstrated satisfactory recognition of implant-supported prostheses and their surrounding key points. The average NE value for key points indicated a high level of accuracy. Statistical studies confirmed no significant difference in the crown-implant ratio between machine and manual measurement results (P>.05).
Conclusions:
Machine learning proved effective in identifying implant-supported prostheses and detecting their crown-implant ratios. If applied as a clinical tool for analyzing radiographs, this research can assist dentists in efficiently and accurately obtaining crown-implant ratio results.
