Automated facial landmark measurement using machine learning: A feasibility study.

Area of Science:

• Dental sciences
• Computer vision
• Anthropometry

Background:

• Facial landmark measurement in prosthodontics often relies on manual techniques.
• Limited information exists on applying machine learning (ML) for these measurements.

Purpose of the Study:

• To evaluate and compare the reliability, validity, and accuracy of manual versus ML-based facial landmark detection.
• To assess ML techniques for facial anthropological measurements in prosthodontics.

Main Methods:

• Two-dimensional (2D) photographs of 100 individuals were analyzed.
• Facial widths (IPW, LCW, MCW, IAW, ICW) were measured manually and using a convolutional neural network (CNN) model.
• Statistical analyses included ICCs, paired t-tests, Bland-Altman plots, and Pearson correlation.

Main Results:

• Excellent intrarater and interrater reliability (>0.90) were achieved with ML.
• No statistically significant differences were found between manual and ML measurements (P > .05).
• Highly significant positive correlations (P < .001) were observed between both methods for all measurements.

Conclusions:

• Machine learning methods are a reliable and accurate alternative to manual digital techniques for facial anthropometric measurements in prosthodontics.
• ML demonstrates high reliability and validity, supporting its integration into prosthodontic workflows.