Evaluating Guideline Adherence in Gemini-Powered Dental Trauma Workflows: Standalone Gemini Chat vs....

Area of Science:

• Artificial Intelligence in Dentistry
• Clinical Decision Support Systems
• Dental Traumatology Guidelines

Background:

• AI-powered tools offer potential for dental trauma decision support.
• Evaluating the accuracy and consistency of different AI workflows is crucial.
• Existing guidelines from the European Society of Endodontology and International Association of Dental Traumatology provide a basis for clinical management.

Purpose of the Study:

• To compare the accuracy and inter-account consistency of two Google Gemini-powered workflows: standalone Gemini chat and NotebookLM.
• To assess their performance in answering dichotomous clinical questions on traumatized permanent teeth management based on established guidelines.

Main Methods:

• A cross-sectional simulation used 99 dichotomous questions derived from ESE and IDTA guidelines.
• Three academic endodontists submitted questions to Gemini and NotebookLM via three independent Google accounts.
• Accuracy was measured by agreement with guideline answers; consistency by identical responses across trials.

Main Results:

• Gemini achieved 83.83% accuracy and 74.74% consistency (κ=0.84).
• NotebookLM demonstrated higher accuracy (92.93%) and perfect consistency (100%, κ=1.00).
• NotebookLM's superior consistency was statistically significant (p < 0.001), though accuracy differences were not significant (p=0.076).

Conclusions:

• Document grounding, as implemented in NotebookLM, enhances AI repeatability and alignment with guideline-derived decision points for dichotomous inquiries.
• NotebookLM achieved complete inter-account consistency and improved accuracy, highlighting the benefits of grounded AI workflows.
• While promising, these workflow-level results require professional oversight and further validation before clinical application.