Automated landmark detection on lateral photographs using artificial intelligence: diagnostic accuracy compared to...

Area of Science:

• Orthodontics
• Medical Imaging
• Artificial Intelligence

Background:

• Lateral cephalograms are crucial for orthodontic diagnostics and treatment planning.
• Soft tissue landmark detection on these images is essential for facial analysis but suffers from human annotator variability.
• Developing automated methods for landmark detection is vital to improve accuracy and consistency.

Purpose of the Study:

• To train an artificial intelligence (AI) algorithm for automated landmark detection on lateral photographs.
• To create a high-quality gold standard dataset for evaluating landmark detection accuracy.
• To compare the performance of the AI algorithm against clinical experts.

Main Methods:

• An AI algorithm was trained on 991 lateral photographs with 14 soft tissue landmarks annotated by experts.
• A separate dataset of 56 photographs was annotated by 11 experts to establish a gold standard.
• AI and expert performance were compared against the gold standard using anthropometric measurements and statistical tests.

Main Results:

• The AI model achieved over 95% detection accuracy for 12 out of 14 landmarks at a 2.0 mm threshold.
• AI predictions exhibited lower variability and mean radial errors than individual expert annotations, especially for landmarks with high inter-annotator disagreement.
• Anthropometric measurements derived from AI predictions showed smaller absolute errors compared to expert-derived measurements.

Conclusions:

• AI-based landmark detection on lateral photographs offers accuracy comparable to expert annotations.
• The AI approach provides greater consistency, particularly for landmarks with significant inter-annotator variability.
• This technology has the potential to enhance the reliability of orthodontic diagnostics and treatment planning.