Forecasting endodontic outcomes from intraoral radiographs using a deep convolutional neural network
Odontology · 16.11.2026
DOI: 10.1007/s10266-026-01616-9 ↗Kurz gefasst
Fachartikel in der Zeitschrift Odontology, online erschienen am 16.11.2026. Verfasst von Ukseong Kim und 4 weiteren Autor:innen. Beteiligte Einrichtung: Microscope Center, Department of Conservative Dentistry and Oral Science Research Center, Yonsei University College of Dentistry, 50-1 Yonsei-Ro, Seo…. Die Arbeit nennt 21 Literaturangaben. Gefördert von: Yonsei University College of Dentistry.
Abstract (Originalsprache)
This study aimed to develop and evaluate a generalized deep convolutional neural network (DCNN) model, ApexNet-17, for predicting endodontic outcomes across all tooth types using preoperative periapical radiographs. A dataset comprising 2,603 teeth from two institutions was utilized. Endodontic outcomes were determined based on tooth survival and radiographic periapical health during 1–3 years of follow-up using the periapical index (PAI). A 17-layer DCNN incorporating self-attention, attention pooling, and five radiographic features was trained and internally validated using 2,279 cases and externally tested using 324 cases. Gradient-weighted class activation mapping (Grad-CAM) was applied to visualize regions contributing to model predictions. ApexNet-17 achieved an accuracy of 72.1%, a sensitivity of 58.5%, and an AUC of 0.666—performance that, while improved over the pilot model, remains below the level typically considered acceptable for clinical decision support. Compared with the previous premolar-only model, ApexNet-17 showed statistically higher accuracy and sensitivity (p < 0.05) but substantially lower specificity (50.6% vs. 78.3%), indicating a higher false-positive rate. Because the models were evaluated using different datasets, these comparisons should be interpreted cautiously, and statistical significance should not be regarded as evidence of clinical superiority. Limitations include the exclusion of intermediate PAI scores (2 and 3), the number and characteristics of which could not be assessed retrospectively, and the uncertain performance of the model in clinically ambiguous cases. Further refinement and validation are required before clinical implementation, although the findings support the feasibility of AI-based prediction of endodontic outcomes using preoperative periapical radiographs.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Microscope Center, Department of Conservative Dentistry and Oral Science Research Center, Yonsei University College of Dentistry, 50-1 Yonsei-Ro, Seodaemun-Gu, Seoul, 03722, Korea.
- Department of Oral and Maxillofacial Radiology, Yonsei University College of Dentistry, 50-1 Yonsei-Ro, Seodaemun-Gu, Seoul, 03722, Korea.
- Microscope Center, Department of Conservative Dentistry and Oral Science Research Center, Yonsei University College of Dentistry, 50-1 Yonsei-Ro, Seodaemun-Gu, Seoul, 03722, Korea. seone1@yuhs.ac.
ORCID-Profile: Ukseong Kim, Euiseong Kim, Chena Lee
Förderung
- Yonsei University College of Dentistry
Zitieren
APA
Kim, U., Hwang, J., Kim, E., Lee, C., & Kim, S. (2026). Forecasting endodontic outcomes from intraoral radiographs using a deep convolutional neural network. Odontology. https://doi.org/10.1007/s10266-026-01616-9
Vancouver
Kim U, Hwang J, Kim E, Lee C, Kim S. Forecasting endodontic outcomes from intraoral radiographs using a deep convolutional neural network. Odontology. 2026. doi:10.1007/s10266-026-01616-9
BibTeX
@article{Kim2026,
author = {Kim, Ukseong and Hwang, Junghwan and Kim, Euiseong and Lee, Chena and Kim, Sunil},
title = {Forecasting endodontic outcomes from intraoral radiographs using a deep convolutional neural network},
journal = {Odontology},
year = {2026},
doi = {10.1007/s10266-026-01616-9}
}
EUVIMED