AI in Learning Anatomy and Restoring Central Incisors: A Comparative Study.

Area of Science:

• Dental Science
• Computer Science
• Artificial Intelligence

Background:

• Dental trauma affects over 1 billion people globally, impacting oral health and aesthetics, especially in children.
• Current restorations involve emergency composite resin work and lab-fabricated mock-ups.
• Advancements in artificial intelligence (AI) offer new possibilities for dental restoration procedures.

Purpose of the Study:

• To compare the performance of different AI approaches for learning and reconstructing central incisor anatomy.
• To evaluate the accuracy and explainability of Principal Component Analysis (PCA) and Deep Learning of Signed Distance Functions (DeepSDF) models.

Main Methods:

• Collected STL files of mature permanent maxillary incisors from three universities.
• Trained PCA and DeepSDF models, performing 3-fold cross-validation.
• Assessed reconstruction accuracy using metrics like surface/volume differences, lengths, Euclidean/Hausdorff distances, and crown-root angulations.
• Evaluated explainability using feature contribution analysis (PCA) and t-SNE (DeepSDF).

Main Results:

• DeepSDF showed significantly higher precision in surface, volume, and Hausdorff distance metrics compared to PCA.
• DeepSDF's reconstruction performance varied with latent code size, with larger sizes showing better results.
• DeepSDF raised concerns regarding model explainability, while PCA's explainability was assessed via feature contribution analysis.

Conclusions:

• Both PCA and DeepSDF show potential for learning and reconstructing central incisor anatomy, with DeepSDF exhibiting superior accuracy.
• Future research should enhance DeepSDF's explainability and incorporate broader smile design factors for clinical trust.
• The high performance of these AI approaches suggests potential clinical applications in smile design and oral rehabilitation.