Related Experiment Video
Updated: May 30, 2025
Digital Hybrid Model Preparation for Virtual Planning of Reconstructive Dentoalveolar Surgical Procedures
Published on: August 5, 2021
3D face reconstruction for maxillofacial surgery based on morphable models and neural networks: A preliminary
Zhewei Chen1, Binghang Li2, Bowen Lei1
1Department of Cranio-Maxillofacial Surgery, Plastic Surgery Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, People's Republic of China.
Objectives:
This study aimed to evaluate the anthropometric accuracy of 3D face reconstruction based on neural networks (3DFRBN) using 2D images, including the assessment of global errors and landmarks, as well as linear and angular measurements.
Methods:
Thirty healthy volunteers were recruited in this study. For each volunteer, five standard photos were taken, capturing anterior, 45° to left and right, and 90° to left and right views. These 2D images were then imported into the 3DFRBN program for generating 3D facial models. Accuracy of landmarks on these models was analyzed in comparison with 3dMD-captured models, as well as global error, and linear and angular measurements.
Results:
The average root mean square error of the 3DFRBN-generated models was 1.52 ± 0.23 mm. Euclidean distance for landmark deviation on 3DFRBN-generated models was 1.87 ± 0.13 mm, ranging from 1.21 mm to 2.17 mm. 13 out of 16 measurements had ICCs over 0.90. The average errors for linear and angular measurements were 0.67 mm and 3.25°, respectively.
Conclusions:
This study introduced a novel, convenient, and accurate approach for both prospective and retrospective 3D facial data acquirement. Its great potential for retrospective 3D facial data acquisition can greatly benefit trauma and oncological reconstruction, as well as retrospective studies.
