The technical quality of root canal filling performed by undergraduate students in pre-clinical education: Instructor versus ChatGPT-4o assessment

Purpose: The present study aims to compare the radiographic technical quality of root canal fillings performed on 3D-printed resin and extracted natural teeth, and to examine the agreement between endodontist educator assessment and analysis performed by an artificial intelligence-based chatbot (ChatGPT-4o).
Methods: After theoretical training and practical demonstration, 3rd-year undergraduate students performed root canal treatment on 108 printed resin teeth and 108 extracted natural teeth. In the radio-graphic examination, parameters such as root canal filling length, filling continuity, apical transport, perforation, and instrument fracture were evaluated by both an experienced instructor and ChatGPT-4o. The technical quality of root canal filling and all procedural errors were compared between the printed teeth and extracted teeth groups using the chi-square test. Interexaminer reliability was measured between the instructor and ChatGPT-4o.
Results: Regarding overall root canal quality, 34% of extracted teeth were acceptable and 66% were unacceptable, while 45% of printed teeth were acceptable and 55% were unacceptable. There is no statistically significant difference between the acceptability rates of extracted teeth vs printed teeth (p>0.05). It was observed that the extracted teeth had more under-filled canals and fewer adequately filled canals than expected, whereas printed teeth were more likely to be adequately filled (p<0.05). There was no difference between sample types having adequate or inadequate filling continuity (p>0.05). Apical transportation, perforation, and instrument fracture rates did not differ significantly between extracted and printed teeth. Cohen’s kappa value is 0.210, and the inter-observer agreement was 62%. These results indicated low agreement between the instructor and ChtaGPT-4o significantly (p<0.05).
Conclusion: The overall quality of the canal filling applied by undergraduate students on 3D-printed resin and extracted teeth was similar. ChatGPT-4o evaluation did not demonstrate a high level of agreement with the endodontist instructor.

Keywords: 3D printed teeth, artificial intelligence, preclinical education.