Evaluating AI Models in Dental Education: Potential for Learning and Clinical Training

Abstract

Artificial intelligence enhances efficiency and accuracy in clinical dentistry, and deep learning (DL) has achieved significant advancements in the field. However, its application in dental education remains underdeveloped. Dental education involves demanding practical training and assessments. Traditional visual-tactile methods are time-consuming, subjective and prone to inconsistencies.

This thesis investigated the applicability of DL tools for assessing tooth cavity preparation and intraoral radiographic techniques.

Three studies were conducted: (1) intraoral images captured with a commercial intraoral camera were used to assess damage to adjacent teeth during cavity preparations using a DL pipeline with YOLOv5 for detection and DenseNet-169 for classification; (2) convolutional neural network architectures were applied to student-acquired bitewings (BWs) to detect common positioning errors; (3) large language models (LLMs), ChatGPT o1, o3-mini, Gemini 2.0 and Grok 3, were evaluated in providing feedback on radiographic positioning errors using baseline and engineered prompts.

The DL models assessing intraoral images achieved 0.81 accuracy, outperforming clinical educators with excellent performance in detecting damage requiring restoration. The CNN model for identifying positioning errors in BWs achieved high accuracy: 96.3% for cone cutting, 93.4% for interproximal overlap, and 73.2% for incorrect receptor placement. LLMs showed variable but promising performance. Gemini 2.0 and Grok 3 performed best for interproximal overlap with baseline prompts, while prompt-engineered ChatGPT o1 and Gemini 2.0 performed better for incorrect film placement.

DL models demonstrated high performance in detecting adjacent-tooth damage in intraoral images and positioning errors on BWs. Open-source LLMs showed variable performance in analysing BW positioning errors. AI-supported assessment is applicable in training dental procedures involving intraoral and radiographic images.