Why This Matters
Panoramic radiographs remain essential in planning third molar extractions, but interpretation requires skill and context. With patients increasingly turning to AI for instant health answers, researchers at Universidad Europea de Madrid set out to test whether ChatGPT-4o—a multimodal large language model—could accurately interpret lower third molar panoramic radiographs.
How the Study Worked
30 anonymized panoramic radiographs showing lower third molars were used.
Each image was paired with a patient-style prompt:
“I have had an x-ray to look at my lower wisdom teeth in case I need to have them removed. In a short one-paragraph answer, tell me what you see.”
900 total responses were generated (30 per radiograph).
Two oral surgery experts scored each as correct, partially correct/incomplete, or incorrect.
A third expert resolved disagreements.
The Results
Overall accuracy: 38.44%
Repeatability: 82.7% agreement in repeated responses, but only moderate agreement quality (Gwet’s AC 60.4%).
ChatGPT often identified implants, restorations, and orthodontic appliances correctly.
It struggled with molar angulation, eruption stage, and overlap with adjacent structures.
Occasional “hallucinations” — fabricated findings not supported by the image — were observed.
Why It’s Different
Unlike most AI-in-dentistry studies, this one used a natural, patient-style question instead of technical prompts. This more closely reflects how patients might use AI for quick guidance—and highlights the risk of them receiving inaccurate interpretations without professional input.
How It Compares to Other Research
ChatGPT’s performance here mirrors other radiology studies, where diagnostic accuracy typically falls between 35% and 50%. Purpose-built convolutional neural networks (CNNs) have achieved higher accuracy for narrow diagnostic tasks, but they require large, well-labeled dental datasets—something still in short supply.
Implications for Dental Practices
For now, ChatGPT-4o should be seen as a patient communication tool, not a diagnostic authority. It can help explain concepts in plain language, but cannot replace the trained eye of a dentist in reading panoramic radiographs—especially for nuanced surgical planning.
Bottom Line
ChatGPT may one day assist in interpreting panoramic radiographs (and the recent release of version 5 will be interesting to evaluate), but today’s models aren’t accurate enough to guide treatment independently. Dentists should be prepared to both harness and supervise AI—ensuring it supports, not supplants, clinical expertise.
Make sure you stay updated on the latest in the dental industry by subscribing