EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Development of a YOLOv8-based deep learning framework for detection and classification of periapical radiolucencies in dental radiographs using the periapical index

Natdanai Hirata, Wannakamon Panyarak, Kittipit Klanliang, Arnon Charuakkra, Kittichai Wantanajittikul

Journal of Associated Medical Sciences · 23.03.2026 · 59 (2) · 316-326

DOI: 10.66285/jams.2026.063 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Associated Medical Sciences, online erschienen am 23.03.2026. Verfasst von Natdanai Hirata und 4 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref). Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

Apical periodontitis (AP) is one of the most common oral diseases worldwide. Radiographs are widely used for diagnosing AP. However, their interpretation is subjective and time-consuming, contributing to increased clinical workload. Although deep learning has been introduced to assist diagnosis, most existing models are trained on limited sets of teeth, restricting their generalizability.

Objective

To investigate the potential of the YOLOv8 model for detecting and classifying periapical radiolucencies in periapical radiographs using the PAI criteria. Materials and methods: A total of 1,191 periapical radiographs from both mandibular and maxillary jaws were collected and scored. Experts annotated 5,498 bounding boxes covering periapical root areas (PRAs). YOLOv8 models were developed to detect PRAs and classify PAI scores, which were grouped into three classes: PAI 1, PAI 2, and PAI 3-5. A custom post-processing method was implemented to improve inference by filtering overlapping detections. Filtered predictions from the test dataset were evaluated for both post-processing methods using precision, recall, F1-score, and mean average precision at an IoU threshold of 0.5 (mAP50). Confusion matrices were used to analyze misclassification patterns, while average IoU assessed overall detection performance.

Results

The best YOLOv8 model achieved a mean average precision at an intersection over union (IoU) threshold of 0.5 (mAP50) of 0.59 with a global average IoU of 0.62. A binary classification ablation model, grouping PAI 1-2 as “healthy” and PAI 3–5 as “diseased,” achieved an improved mAP50 of 0.66. The custom post-processing algorithm successfully eliminated all overlapping boxes and enhanced model performance.

Conclusions

Despite moderate performance, the YOLOv8-based models demonstrate potential as adjunct tools to assist dental practitioners in radiographic evaluation, particularly for less experienced clinicians. Further development is required to improve clinical applicability and robustness.

Zitieren

APA

Hirata, N., Panyarak, W., Klanliang, K., Charuakkra, A., & Wantanajittikul, K. (2026). Development of a YOLOv8-based deep learning framework for detection and classification of periapical radiolucencies in dental radiographs using the periapical index. Journal of Associated Medical Sciences, 59(2), 316-326. https://doi.org/10.66285/jams.2026.063

Vancouver

Hirata N, Panyarak W, Klanliang K, Charuakkra A, Wantanajittikul K. Development of a YOLOv8-based deep learning framework for detection and classification of periapical radiolucencies in dental radiographs using the periapical index. Journal of Associated Medical Sciences. 2026;59(2):316-326. doi:10.66285/jams.2026.063

BibTeX
@article{Hirata2026,
  author = {Hirata, Natdanai and Panyarak, Wannakamon and Klanliang, Kittipit and Charuakkra, Arnon and Wantanajittikul, Kittichai},
  title = {Development of a YOLOv8-based deep learning framework for detection and classification of periapical radiolucencies in dental radiographs using the periapical index},
  journal = {Journal of Associated Medical Sciences},
  year = {2026},
  volume = {59},
  number = {2},
  pages = {316--326},
  doi = {10.66285/jams.2026.063}
}
Kostenlos starten →