Artificial Intelligence in Caries Risk Assessment: Evaluating the Current Status of CAMBRA and Cariogram with Large...

Area of Science:

• Artificial Intelligence in Healthcare
• Dental Public Health
• Pediatric Dentistry

Background:

• Caries risk assessment (CRA) is crucial for personalized early childhood caries management.
• Standardizing existing CRA tools in clinical practice is challenging.
• The efficacy of large language model (LLM)-based AI in applying CRA tools is not well understood.

Purpose of the Study:

• To evaluate ChatGPT and Google Gemini's ability to apply and interpret early childhood CRA tools.
• To assess AI performance under different prompting conditions (unguided, guideline-informed, guideline-only).

Main Methods:

• Thirty pediatric clinical vignettes (0-5 years) were analyzed using CAMBRA and Cariogram tools.
• ChatGPT and Google Gemini were prompted under three conditions.
• AI outputs were compared against expert classifications using categorical agreement, mean absolute error (MAE), quality, accuracy, and readability metrics.

Main Results:

• Guideline-informed and guideline-only prompting significantly improved AI performance (reduced MAE, enhanced quality and accuracy) compared to unguided prompting.
• ChatGPT demonstrated lower MAE than Gemini for the Cariogram tool.
• AI performance varied depending on the specific CRA tool and prompting strategy.

Conclusions:

• LLM-based AI systems show potential for supporting early childhood caries risk assessment, especially with guideline-based prompting.
• The effectiveness of AI in CRA is influenced by the specific AI model and the complexity of the CRA tool.
• Numerical or algorithmic aspects of tools like Cariogram present challenges for current LLM-based AI systems.