Edentulous space classification and planning of clasp-retained removable dentures using generative artificial...

Area of Science:

• Dental Prosthetics
• Artificial Intelligence in Healthcare
• Diagnostic Tools

Background:

• Partially dentate jaws present challenges in prosthetic treatment planning.
• Accurate classification of edentulous areas is crucial for designing removable dentures.
• Generative artificial intelligence (AI) offers potential for automating diagnostic tasks.

Purpose of the Study:

• To evaluate generative AI's ability to identify and classify partially dentate jaws using Kennedy's system.
• To assess AI's capability in planning clasp-retained removable dentures.
• To compare the performance of different generative AI models (GPTs and o1-preview) in these tasks.

Main Methods:

• One hundred anonymized partially dentate jaw datasets were used.
• OpenAI's GPTs (GPT-4o model) and o1-preview were trained and prompted for gap analysis and Kennedy classification.
• AI-generated prosthetic plans were compared against a dentist's plan using statistical analyses (Cohen's kappa, McNemar's test, Chi-Square Test).

Main Results:

• o1-preview significantly outperformed GPTs in identifying gaps (83% vs. 14%) and classifying Kennedy classes (90% vs. 40%).
• Both AI models showed very low agreement (1-6%) with the dentist's treatment plan for denture design.
• o1-preview made fewer and different types of errors compared to GPTs.

Conclusions:

• Generative AI, particularly o1-preview, demonstrates promise for dental diagnostic tasks like gap identification and classification.
• Current AI models are not yet proficient in reliably planning clasp-retained removable dentures.
• Integration into clinical workflows requires automated solutions and enhanced prompt engineering for AI in dentistry.