Automatic detection of orthodontically induced external root resorption based on deep convolutional neural networks...

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Orthodontically-induced external root resorption (OIERR) is a frequent complication of orthodontic treatment.
• Current diagnostic methods for OIERR rely on subjective assessments and time-consuming manual measurements.
• There is a need for objective and efficient tools to aid in OIERR diagnosis.

Purpose of the Study:

• To develop and evaluate an intelligent detection model for OIERR using deep convolutional neural networks (CNNs).
• To leverage cone-beam computed tomography (CBCT) images for automated OIERR detection.
• To provide auxiliary diagnostic support for orthodontists.

Main Methods:

• Six pretrained CNN architectures were employed to construct OIERR detection models.
• A dataset of 1717 CBCT slices was used for training, and 429 slices for testing.
• Model performance was evaluated using accuracy, precision, sensitivity, specificity, and F1-score, and compared to human expert diagnoses.

Main Results:

• The EfficientNet-B1 model demonstrated superior performance in detecting OIERR.
• The model achieved high metrics: 0.97 accuracy, 0.98 precision, 0.97 sensitivity, 0.98 specificity, and 0.98 F1-score.
• The CNN model significantly outperformed two orthodontists in accuracy (0.97 vs. 0.86), recall (0.97 vs. 0.78), and F1-score (0.98 vs. 0.87).

Conclusions:

• Deep convolutional neural networks (CNNs) combined with CBCT imaging offer an accurate and efficient method for OIERR detection.
• The developed AI model surpasses the diagnostic capabilities of human orthodontists.
• This AI-driven approach is poised to become a valuable clinical tool for rapid screening and diagnosis of OIERR.