Artificial intelligence for caries detection: a novel diagnostic tool using deep learning algorithms.

OBJECTIVES: The aim of this study was to develop an assessment tool for automatic detection of dental caries in periapical radiographs using convolutional neural network (CNN) architecture. METHODS: A novel diagnostic model named ResNet + SAM was established using numerous periapical radiographs (4278 images) annotated by medical experts to automatically detect dental caries. The performance of the model was compared to the traditional CNNs (VGG19, ResNet-50), and the dentists. The Gradient-weighted Class Activation Mapping (Grad-CAM) technique shows the region of interest in the image for the CNNs. RESULTS: ResNet + SAM demonstrated significantly improved performance compared to the modified ResNet-50 model, with an average F1 score of 0.886 (95% CI 0.855-0.918), accuracy of 0.885 (95% CI 0.862-0.901) and AUC of 0.954 (95% CI 0.924-0.980). The comparison between the performance of the model and the dentists revealed that the model achieved higher accuracy than that of the junior dentists. With the assist of the tool, the dentists achieved superior metrics with a mean F1 score of 0.827 and the interobserver agreement for dental caries is enhanced from 0.592/0.610 to 0.706/0.723. CONCLUSIONS: According to the results obtained from the experiments, the automatic assessment tool using the ResNet + SAM model shows remarkable performance and has excellent possibilities in identifying dental caries. The use of the assessment tool in clinical practice can be of great benefit as a clinical decision-making support in dentistry and reduce the workload of dentists.

Artificial Intelligence for Classifying and Archiving Orthodontic Images.

One of the main requirements for orthodontic treatment is continuous image acquisition. However, the conventional system of orthodontic image acquisition, which includes manual classification, archiving, and monitoring, is time-consuming and prone to errors caused by fatigue. This study is aimed at developing an effective artificial intelligence tool for the automated classification and monitoring of orthodontic images. We comprehensively evaluated the ability of a deep learning model based on Deep hidden IDentity (DeepID) features to classify and archive photographs and radiographs. This evaluation was performed using a dataset of >14,000 images encompassing all 14 categories of orthodontic images. Our model automatically classified orthodontic images in an external dataset with an accuracy of 0.994 and macro area under the curve of 1.00 in 0.08 min. This was 236 times faster than a human expert (18.93 min). Furthermore, human experts with deep learning assistance required an average of 8.10 min to classify images in the external dataset, much shorter than 18.93 min. We conclude that deep learning can improve the accuracy, speed, and efficiency of classification, archiving, and monitoring of orthodontic images.

Artificial intelligence-aided detection of ectopic eruption of maxillary first molars based on panoramic radiographs.

OBJECTIVES: Ectopic eruption (EE) of maxillary permanent first molars (PFMs) is among the most frequent ectopic eruption, which leads to premature loss of adjacent primary second molars, impaction of premolars and a decrease in dental arch length. Apart from oral manifestations such asdelayed eruption of PFMs and discoloration of primary second molars, panoramic radiographs can reveal EE of maxillary PFMs as well. Identifying eruption anomalies in radiographs can be strongly experience-dependent, leading us to develop here an automatic model that can aid dentists in this task and allow timelier interventions. METHODS: Panoramic X-ray images from 1480 patients aged 4-9 years old were used to train an auto-screening model. Another 100 panoramic images were used to validate and test the model. RESULTS: The positive and negative predictive values of this auto-screening system were 0.86 and 0.88, respectively, with a specificity of 0.90 and a sensitivity of 0.86. Using the model to aid dentists in detecting EE on the 100 panoramic images led to higher sensitivity and specificity than when three experienced pediatric dentists detected EE manually. CONCLUSIONS: Deep learning-based automatic screening system is useful and promising in the detection EE of maxillary PFMs with relatively high specificity. However, deep learning is not completely accurate in the detection of EE. To minimize the effect of possible false negative diagnosis, regular follow-ups and re-evaluation are required if necessary. CLINICAL SIGNIFICANCE: Identification of EE through a semi-automatic screening model can improve the efficacy and accuracy of clinical diagnosis compared to human experts alone. This method may allow earlier detection and timelier intervention and management.

Artificial intelligence for caries and periapical periodontitis detection.

OBJECTIVES: Periapical periodontitis and caries are common chronic oral diseases affecting most teenagers and adults worldwide. The purpose of this study was to develop an evaluation tool to automatically detect dental caries and periapical periodontitis on periapical radiographs using deep learning. METHODS: A modified deep learning model was developed using a large dataset (4129 images) with high-quality annotations to support the automatic detection of both dental caries and periapical periodontitis. The performance of the model was compared to the classification performance of dentists. RESULTS: The deep learning model automatically distinguished dental caries with an F1-score of 0.829 and periapical periodontitis with an F1-score of 0.828. The comparison of model-only and expert-only detection performance showed that the accuracy of the fully automatic method was significantly higher than that of the young dentists. With deep learning assistance, the experts not only reached a higher diagnostic accuracy with an average F1-score of 0.7844 for dental caries and 0.8208 for periapical periodontitis compared to expert-only scenarios, but also increased inter-observer agreement from 0.585/0.590 to 0.726/0.713 for dental caries and from 0.623/0.563 to 0.752/0.740 for periapical periodontitis. CONCLUSIONS: Based on these experimental results, deep learning can improve the accuracy and consistency of evaluating dental caries and periapical periodontitis on periapical radiographs. CLINICAL SIGNIFICANCE: Deep learning models can improve accuracy and consistency and reduce the workload of dentists, making artificial intelligence a powerful tool for clinical practice.