Designing an artificial intelligence system for dental occlusion classification using intraoral photographs: A comparative analysis between artificial intelligence-based and clinical diagnoses

Introduction

This study aimed to design an artificial intelligence (AI) system for dental occlusion classification using intraoral photographs. Moreover, the performance of this system was compared with that of an expert clinician.

Methods

This study included 948 adult patients with permanent dentition who presented to the Department of Orthodontics, School of Dentistry, Mashhad University of Medical Sciences, during 2022-2023. The intraoral photographs taken from the patients in left, right, and frontal views (3 photographs for each patient) were collected and underwent augmentation, and about 7500 final photographs were obtained. Moreover, the patients were clinically examined by an expert orthodontist for malocclusion, overjet, and overbite and were classified into 6 groups: Class I, Class II, half-cusp Class II, Super Class I, Class III, and unclassifiable. In addition, a multistage neural network system was created and trained using the photographs of 700 patients. Then, it was used to classify the remaining 248 patients using their intraoral photographs. Finally, its performance was compared with that of the expert clinician. All statistical analyses were performed using the Stata software (version 17; Stata Corp, College Station, Tex).

Results

The accuracy, precision, recall, and F1 score of the AI system in the malocclusion classification of molars were calculated to be 93.1%, 88.6%, 91.2%, and 89.7%, respectively, whereas the AI system had an accuracy, precision, recall, and F1 score of 89.1%, 88.8%, 91.42%, and 89.8% for malocclusion classification of canines, respectively. Moreover, the mean absolute error of the AI system accuracy was 1.98 ± 2.11 for overjet and 1.28 ± 1.60 for overbite classifications.

Conclusions

AI exhibited remarkable performance in detecting all classes of malocclusion, which was higher than that of orthodontists, especially in predicting angle classification. However, its performance was not acceptable in overjet and overbite measurement compared with expert orthodontists.