Evaluation of automated cephalometric analysis based on the latest deep learning method.

Area of Science:

• Biomedical Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Automated cephalometric analysis is crucial for orthodontic diagnosis and treatment planning.
• Previous artificial intelligence (AI) methods have shown promise but require further validation.
• Standardized evaluation frameworks, like those from IEEE ISBI challenges, are essential for comparing AI performance.

Purpose of the Study:

• To evaluate a novel deep learning-based AI for automated cephalometric landmark identification.
• To compare its performance against previously published AI methods and human examiners.
• To assess the AI's accuracy using metrics from IEEE ISBI challenges.

Main Methods:

• A deep learning model (modified YOLO version 3) was trained on 1983 cephalograms.
• The AI was tested on 200 cephalograms, identifying 19 cephalometric landmarks.
• Performance was evaluated using Success Detection Rate (SDR) and Success Classification Rate (SCR) against human-identified landmarks.

Main Results:

• The latest AI achieved an SDR of 75.5% within a 2-mm range and an SCR of 81.5%.
• These results surpassed those of previously published AI methods.
• The AI demonstrated a superior SCR compared to human examiners in certain cephalometric measures.

Conclusions:

• The developed AI exhibits superior performance compared to existing AI methods for cephalometric analysis.
• The AI's capabilities in cephalometric analysis appear comparable to those of human experts.
• This advanced AI holds potential for improving the efficiency and accuracy of orthodontic assessments.