Several artificial intelligence-based models have been presented for the detection of periodontal bone loss (PBL), mostly using convolutional neural networks, which are the state of the art in deep learning. Given the emerging breakthrough of transformer networks in computer vision, we aimed to evaluate various models for automatized PBL detection. An image data set of 21,819 anonymized periapical radiographs from the upper/lower and anterior/posterior regions was assessed by calibrated dentists according to PBL. Five vision transformer networks (ViT-base/ViT-large from Google, BEiT-base/BEiT-large from Microsoft, DeiT-base from Facebook/Meta) were utilized and evaluated. Accuracy (ACC), sensitivity (SE), specificity (SP), positive/negative predictive value (PPV/NPV) and area under the ROC curve (AUC) were statistically determined. The overall diagnostic ACC and AUC values ranged from 83.4 to 85.2% and 0.899 to 0.918 for all evaluated transformer networks, respectively. Differences in diagnostic performance were evident for lower (ACC 94.1–96.7%; AUC 0.944–0.970) and upper anterior (86.7–90.2%; 0.948–0.958) and lower (85.6–87.2%; 0.913–0.937) and upper posterior teeth (78.1–81.0%; 0.851–0.875). In this study, only minor differences among the tested networks were detected for PBL detection. To increase the diagnostic performance and to support the clinical use of such networks, further optimisations with larger and manually annotated image data sets are needed.

(1) Background: Caries, periapical lesions, periodontal bone loss (PBL), and endo-perio lesions are common dental findings that require an accurate diagnostic assessment to allow appropriate disease management. The purpose of this reliability study was to compare the inter- and intra-rater reliability for the detection of the above-mentioned pathologies on periapical radiographs. (2) Methods: Fourteen dentists (three with more than two years and eleven with less than two years of work experience) participated in a training workshop prior to data acquisition. A total of 150 radiographs were assessed by all raters in two rounds. Cohen’s Kappa (CK) values and a binary logistic regression were calculated. (3) Results: The reliability was found in a moderate and substantial range of agreement: caries (mean inter-rater CK value/first round 0.704/mean inter-rater CK value/second round 0.659/mean intra-rater CK value 0.778), periapical lesions (0.643/0.611/0.768), PBL (0.454/0.482/0.739) and endo-perio lesion (0.702/0.689/0.840). The regression model revealed a significant influence of the clinical experience, and furthermore, periapical pathologies and PBL were identified less reliably in comparison to caries and endo-perio lesions. (4) Conclusions: The dentist’s ability to detect the chosen pathologies was linked with significant differences. Periapical lesions and PBL were identified less reliably than caries and endo-perio lesions.

Interest in machine learning models and convolutional neural networks (CNNs) for diagnostic purposes is steadily increasing in dentistry. Here, CNNs can potentially help in the classification of periodontal bone loss (PBL). In this study, the diagnostic performance of five CNNs in detecting PBL on periapical radiographs was analyzed. A set of anonymized periapical radiographs (N = 21,819) was evaluated by a group of trained and calibrated dentists and classified into radiographs without PBL or with mild, moderate, or severe PBL. Five CNNs were trained over five epochs. Statistically, diagnostic performance was analyzed using accuracy (ACC), sensitivity (SE), specificity (SP), and area under the receiver operating curve (AUC). Here, overall ACC ranged from 82.0% to 84.8%, SE 88.8–90.7%, SP 66.2–71.2%, and AUC 0.884–0.913, indicating similar diagnostic performance of the five CNNs. Furthermore, performance differences were evident in the individual sextant groups. Here, the highest values were found for the mandibular anterior teeth (ACC 94.9–96.0%) and the lowest values for the maxillary posterior teeth (78.0–80.7%). It can be concluded that automatic assessment of PBL seems to be possible, but that diagnostic accuracy varies depending on the location in the dentition. Future research is needed to improve performance for all tooth groups.

Background/Objectives: Early childhood caries (ECC) is a widespread and severe oral health problem that potentially affects the general health of children. Visual–tactile examination remains the diagnostic method of choice to diagnose ECC, although visual examination could be automated by artificial intelligence (AI) tools in the future. The aim of this study was the external validation of a recently published and freely accessible AI-based model for detecting ECC and classifying carious lesions in dental photographs. Methods: A total of 143 anonymised photographs of anterior deciduous teeth (ECC = 107, controls = 36) were visually evaluated by the dental study group (reference test) and analysed using the AI-based model (test method). Diagnostic performance was determined statistically. Results: ECC detection accuracy was 97.2%. Diagnostic performance varied between carious lesion classes (noncavitated lesions, greyish translucency/microcavity, cavitation, destructed tooth), with accuracies ranging from 88.9% to 98.1%, sensitivities ranging from 68.8% to 98.5% and specificities ranging from 86.1% to 99.4%. The area under the curve ranged from 0.834 to 0.964. Conclusions: The performance of the AI-based model is similar to that reported for the internal dataset used by developers. Further studies with independent image samples are required to comprehensively gauge the performance of the model.

This ex vivo diagnostic study aimed to externally validate an open-access artificial intelligence (AI)-based model for the detection, classification, localisation and segmentation of enamel/molar incisor hypomineralisation (EH/MIH). An independent sample of web images showing teeth with (n = 277) and without (n = 178) EH/MIH was evaluated by a workgroup of dentists whose consensus served as the reference standard. Then, an AI-based model was used for the detection of EH/MIH, followed by automated classification and segmentation of the findings (test method). The accuracy (ACC), sensitivity (SE), specificity (SP) and area under the curve (AUC) were determined. Furthermore, the correctness of EH/MIH lesion localisation and segmentation was evaluated. An overall ACC of 94.3 % was achieved for image-based detection of EH/MIH. Cross-classification of the AI-based class prediction and the reference standard resulted in an agreement of 89.2 % for all diagnostic decisions (n = 594), with an ACC between 91.4 % and 97.8 %. The corresponding SE and SP values ranged from 81.7 % to 92.8 % and 91.9 % to 98.7 %, respectively. The AUC varied between 0.894 and 0.945. Image size had only a limited impact on diagnostic performance. The AI-based model correctly predicted EH/MIH localisation in 97.3 % of cases. For the detected lesions, segmentation was fully correct in 63.4 % of all cases and partially correct in 33.9 %. This study documented the promising diagnostic performance of an open-access AI tool in the detection and classification of EH/MIH in external images. Externally validated AI-based diagnostic methods could facilitate the detection of EH/MIH lesions in dental photographs.

Objective: This ex vivo diagnostic study aimed to externally validate a freely accessible AI-based model for caries detection, classification, localisation and segmentation using an independent image dataset. It was hypothesised that there would be no difference in diagnostic performance compared to previously published internal validation data. Methods: For the independent dataset, 718 dental images representing different stages of carious (n = 535) and noncarious teeth (n = 183) were retrieved from the internet. All photographs were evaluated by the dental team (reference standard) and the AI-based model (test method). Diagnostic performance was statistically determined using cross-tabulations to calculate accuracy (ACC), sensitivity (SE), specificity (SP) and area under the curve (AUC). Results: An overall ACC of 92.0% was achieved for caries detection, with an ACC of 85.5–95.6%, SE of 42.9–93.3%, SP of 82.1–99.4% and AUC of 0.702–0.909 for the classification of caries. Furthermore, 97.0% of the cases were accurately localised. Fully and partially correct segmentation was achieved in 52.9% and 44.1% of the cases, respectively. Conclusions: The validated AI-based model showed promising diagnostic performance in detecting and classifying caries using an independent image dataset. Future studies are needed to investigate the validity, reliability and practicability of AI-based models using dental photographs from different image sources and/or patient groups.