Evaluating the Diagnostic Accuracy of Artificial Intelligence in Periapical Radiographs

Abstract

Background: Over the past few years, significant development has been made in the field of oral and dental diagnostics. A conservative treatment strategy with a favorable prognosis could be implemented by the clinician with an early diagnosis. It has been reported that examiners with greater expertise exhibit more effective diagnostic accuracy than those with less experience. AI is frequently thought of as a useful tool to assist clinicians and dentists in minimizing their workloads.

Objectives: The purpose of this study was to evaluate the diagnostic accuracy of artificial intelligence in identifying common dental problems on periapical radiographs compared with experienced dentists. 

Methods: A total of 283 periapical radiographs were randomly selected from the database of the University Dental Hospital. Two general dentists with more than 10 years of clinical experience manually assessed the periapical radiographs, which was ground truth. The same periapical radiographs were then uploaded into AI dental software.

Results: The obtained Cohen’s Kappa values (0.61-0.8) indicated substantial agreement between the two investigators. Good agreement is noted in several parameters; F1 scores of  apical radiolucency, obturation, and tooth detection were 0.7, 0.9, and 0.8, respectively. For Caries, the model had poor reliability with an accuracy of 61%. 

Conclusion: AI can help detect dental issues on periapical radiographs but still needs improvement before it can fully aid clinical decision making.