Diagnostic Capabilities of Large Language Models in Paediatric Dentistry

Area of Science:

• Artificial Intelligence in Dentistry
• Paediatric Dentistry Applications
• Medical Diagnostic Tools

Background:

• Large language models (LLMs) are increasingly studied for dental diagnosis.
• Existing research often focuses on single topics, limiting understanding of broader clinical utility.
• The application of LLMs in diverse paediatric dental patient scenarios remains underexplored.

Purpose of the Study:

• To evaluate the diagnostic accuracy of two LLMs (Copilot and Gemini) in paediatric dentistry.
• To assess the impact of including or omitting patient chief complaints on LLM diagnostic performance.
• To identify specific areas where LLMs excel or struggle in paediatric dental diagnosis.

Main Methods:

• Twenty-five paediatric dentistry cases from a national examination were selected.
• Patient data (age, sex, chief complaint) were presented to LLMs alongside intraoral photos and radiographs.
• LLMs were tasked with identifying abnormal areas and providing diagnoses under varying information conditions.

Main Results:

• Both LLMs demonstrated >70% accuracy in identifying pathological locations and ~60% accuracy in diagnosis.
• Diagnostic accuracy significantly decreased when the chief complaint was omitted.
• LLM performance faltered on paediatric-specific issues like tooth number abnormalities and eruption disorders without a chief complaint.

Conclusions:

• LLMs possess potential for paediatric dentistry, but accurate chief complaints are crucial for optimal performance.
• LLMs currently exhibit limitations in diagnosing complex, paediatric-specific dental conditions.
• Further development is needed to enhance LLM capabilities in specialized paediatric dentistry.