Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2024 Mar 26;24(1):387.
doi: 10.1186/s12903-024-04129-5.

Combining public datasets for automated tooth assessment in panoramic radiographs

Niels van Nistelrooij  1   2 , Khalid El Ghoul  3 , Tong Xi  4 , Anindo Saha  4 , Steven Kempers  1 , Max Cenci  5 , Bas Loomans  5 , Tabea Flügge  6   7 , Bram van Ginneken  4 , Shankeeth Vinayahalingam  1
Affiliations

Combining public datasets for automated tooth assessment in panoramic radiographs

Niels van Nistelrooij et al. BMC Oral Health. .

Abstract

Objective: Panoramic radiographs (PRs) provide a comprehensive view of the oral and maxillofacial region and are used routinely to assess dental and osseous pathologies. Artificial intelligence (AI) can be used to improve the diagnostic accuracy of PRs compared to bitewings and periapical radiographs. This study aimed to evaluate the advantages and challenges of using publicly available datasets in dental AI research, focusing on solving the novel task of predicting tooth segmentations, FDI numbers, and tooth diagnoses, simultaneously.

Materials and methods: Datasets from the OdontoAI platform (tooth instance segmentations) and the DENTEX challenge (tooth bounding boxes with associated diagnoses) were combined to develop a two-stage AI model. The first stage implemented tooth instance segmentation with FDI numbering and extracted regions of interest around each tooth segmentation, whereafter the second stage implemented multi-label classification to detect dental caries, impacted teeth, and periapical lesions in PRs. The performance of the automated tooth segmentation algorithm was evaluated using a free-response receiver-operating-characteristics (FROC) curve and mean average precision (mAP) metrics. The diagnostic accuracy of detection and classification of dental pathology was evaluated with ROC curves and F1 and AUC metrics.

Results: The two-stage AI model achieved high accuracy in tooth segmentations with a FROC score of 0.988 and a mAP of 0.848. High accuracy was also achieved in the diagnostic classification of impacted teeth (F1 = 0.901, AUC = 0.996), whereas moderate accuracy was achieved in the diagnostic classification of deep caries (F1 = 0.683, AUC = 0.960), early caries (F1 = 0.662, AUC = 0.881), and periapical lesions (F1 = 0.603, AUC = 0.974). The model's performance correlated positively with the quality of annotations in the used public datasets. Selected samples from the DENTEX dataset revealed cases of missing (false-negative) and incorrect (false-positive) diagnoses, which negatively influenced the performance of the AI model.

Conclusions: The use and pooling of public datasets in dental AI research can significantly accelerate the development of new AI models and enable fast exploration of novel tasks. However, standardized quality assurance is essential before using the datasets to ensure reliable outcomes and limit potential biases.

Keywords: Artificial intelligence; Diagnostic classification; Panoramic radiograph; Public datasets; Tooth segmentation.

PubMed Disclaimer

Conflict of interest statement

The authors declare that they have no competing interests.

Figures

Fig. 1
Fig. 1
Overview of methodology. The teeth in the input PR (a) are segmented by Mask DINO and their FDI numbers are predicted(b) [25]. For each predicted tooth, a cropped image is made with the segmentation as extra channel (c, ROI extraction). This image is processed by four binary classifiers, one for each diagnosis, whose predictions are aggregated using an MLP [34], and by a multi-label classifier who returns multiple predictions via its CSRA heads [40]. All multi-label predictions are summed and these final scores are used to add diagnoses to the tooth segmentations (d). Non-diagnosed teeth are predicted, but not shown in the result for clarity. The label is the tooth’s FDI number with C = early caries, D = deep caries, P = periapical lesion, I = impacted
Fig. 2
Fig. 2
Free-response receiver operating characteristic (FROC) curves of tooth instance segmentation on PRs. Results are based on 324 held-out test PRs from the OdontoAI platform. The score is computed as the mean sensitivity at false-positive rates of 0.25, 0.5, 1, 2, 4, and 8 following [48]
Fig. 3
Fig. 3
Tooth segmentation failure cases. Two PRs from the DENTEX challenge dataset are shown with predictions. A horizontally impacted canine (in red circle) is not segmented in the first PR and multiple teeth (e.g. in red circle) are not segmented in the second PR due to a syndromic disease. The label is the tooth’s FDI number
Fig. 4
Fig. 4
Receiver operating characteristic (ROC) curves illustrating the multi-label classification results of tooth diagnoses on PRs. A varying effectiveness can be observed for different tooth diagnoses. Results are based on 134 held-out test PRs from the DENTEX challenge. AUC = area under ROC curve
Fig. 5
Fig. 5
Confusion matrix illustrating the multi-label classification results of tooth diagnoses on PRs. Results on 134 held-out test PRs from the DENTEX challenge are shown for the most effective model. The colorbar is normalized according to the number of PRs per predicted label
Fig. 6
Fig. 6
DENTEX annotations with missing diagnoses. A missed diagnosis (false negative) is shown for each diagnosis, which are selected based on the maximum difference between a diagnosis probability and whether that diagnosis is annotated. Note that the DENTEX dataset only annotates diagnosed teeth, whereas the current method predicts all teeth. The label is the tooth’s FDI number with C = early caries, D = deep caries, P = periapical lesion, I = impacted
Fig. 7
Fig. 7
Misdiagnosed DENTEX annotations. A misdiagnosis (false positive) is shown for each diagnosis, which are selected based on the maximum difference between a diagnosis probability and whether that diagnosis is annotated. Note that the DENTEX dataset only annotates diagnosed teeth, whereas the current method predicts all teeth. The label is the tooth’s FDI number with C = early caries, D = deep caries, P = periapical lesion, I = impacted
See this image and copyright information in PMC

Similar articles

See all similar articles

References

    1. FelsyPremila G, et al. Visual interpretation of panoramic radiographs in dental students using eye-tracking technology. J Dent Educ. Mar. 2022;86. 10.1002/jdd.12899. - PubMed
    1. Peretz B, Gotler M, Kaffe I. Common Errors in Digital Panoramic Radiographs of Patients with Mixed Dentition and Patients with Permanent Dentition. In: International journal of dentistry 2012Feb. (2012), p. 584138. 10.1155/2012/584138. - PMC - PubMed
    1. Susanne Perschbacher. Interpretation of panoramic radiographs. In: Australian dental journal 57 Suppl 1 (Mar. 2012), pp. 40–5. 10.1111/j.1834-7819.2011.01655.x. - PubMed
    1. Lei C, et al. Expert consensus on dental caries management. Int J Oral Sci. Dec. 2022;14. 10.1038/s41368-022-00167-3. - PMC - PubMed
    1. Akarslan ZZ et al. A comparison of the diagnostic accuracy of bitewing, periapical, unfiltered and filtered digital panoramic images for approximal caries detection in posterior teeth. In: Dentomaxillofacial Radiology 37.8 (2008). PMID: 19033431, pp. 458–463. 10.1259 / dmfr/84698143. - PubMed