Orthodontic craniofacial pattern diagnosis: cephalometric geometry and machine learning.

Area of Science:

• Orthodontics
• Medical Imaging
• Machine Learning

Background:

• Accurate craniofacial pattern diagnosis is essential for effective orthodontic treatment.
• Machine learning (ML) offers potential for high-precision, time-saving diagnostic tools, but requires clinical validation.

Purpose of the Study:

• To develop and validate a craniofacial ML diagnostic workflow based on a cephalometric geometric model.
• To assess the reliability and consistency of ML algorithms in diagnosing sagittal and vertical skeletal patterns against clinical norms.

Main Methods:

• A cephalometric geometric model was created from 408 lateral cephalograms to identify landmark locations.
• Nine supervised ML algorithms were applied to geometric features for classifying sagittal and vertical skeletal patterns.
• Dimension reduction and visualization techniques (decision boundaries, landmark contribution contours) were used to analyze diagnostic consistency.

Main Results:

• Multi-layer perceptron achieved 97.56% accuracy for sagittal patterns; linear support vector machine reached 90.24% for vertical patterns.
• Sagittal diagnoses (91.60 ± 5.43%) were generally superior to vertical diagnoses (82.25 ± 6.37%).
• Discriminative ML algorithms showed more stable performance (93.20 ± 3.29%) than generative ones (85.98 ± 9.48%).

Conclusions:

• The proposed craniofacial ML workflow demonstrates high consistency with clinical diagnostic norms.
• This ML approach can serve as a valuable supplement to traditional orthodontic diagnosis.
• The study highlights the potential of validated ML models in improving craniofacial diagnostics.