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

Automating the prognosis of unicystic ameloblastoma on haematoxylin and eosin multi- resolution whole slide images using artificial intelligence

K. Sowmya, Dhanya G. Athikary, Harsh Tiwari, Harshit Kumar, Divya Biligere Shivanna, Roopa S. Rao, Madhusudan Astekar

Digital Dentistry Journal · 2026-12 · 4 (1) · 100107

DOI: 10.1016/j.ddj.2026.100107 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Digital Dentistry Journal, 2026-12. Verfasst von K. Sowmya und 6 weiteren Autor:innen. Die Arbeit nennt 14 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Vision Group on Science and Technology.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

Unicystic ameloblastoma (UA) is a biologically distinct odontogenic tumour characterised by complex histopathological features and a high recurrence rate. Accurate prediction of recurrence is essential for optimising treatment decisions and improving long-term patient outcomes.

Objective

To develop and validate a deep learning model, the Cross-Resolution Network (CRN) for automated prediction of recurrence in UA, utilising Hematoxylin and Eosin (H&E)-stained whole-slide images (WSIs) at 20x and 40x magnifications.

Methods

A retrospective dataset comprising annotated WSIs of confirmed UA cases was curated and augmented using advanced image processing techniques. The proposed CRN architecture was designed to integrate multi-resolution inputs through resolution-specific embeddings, capturing both fine-grained and contextual morphological features. Model performance was evaluated and compared to a benchmark deep learning model, ResCRAMNet.

Results

The CRN outperformed the ResCRAMNet model, achieving a recurrence prediction accuracy of 96%. It demonstrated superior sensitivity and precision in identifying histological patterns associated with recurrence, due to its ability to synthesise features across resolutions. Clinical Application The CRN model can be integrated into digital pathology workflows to assist oral pathologists in early and accurate recurrence risk stratification. It supports personalised treatment planning by informing decisions on conservative versus radical surgical approaches and enables risk-based follow-up protocols. This AI-enabled tool enhances diagnostic confidence, reduces the likelihood of recurrence through tailored management, and improves overall patient care in oral pathology.

Conclusion

The Cross-Resolution Network offers a robust and clinically relevant AI framework for recurrence prediction in UA. By leveraging multi-resolution histopathological features, it provides a valuable adjunct to conventional diagnostic processes, with the potential to significantly impact prognostication and treatment strategies.

Förderung

  • Vision Group on Science and Technology

Zitieren

APA

Sowmya, K., Athikary, D. G., Tiwari, H., Kumar, H., Shivanna, D. B., Rao, R. S., & Astekar, M. (2026). Automating the prognosis of unicystic ameloblastoma on haematoxylin and eosin multi- resolution whole slide images using artificial intelligence. Digital Dentistry Journal, 4(1), 100107. https://doi.org/10.1016/j.ddj.2026.100107

Vancouver

Sowmya K, Athikary DG, Tiwari H, Kumar H, Shivanna DB, Rao RS, et al. Automating the prognosis of unicystic ameloblastoma on haematoxylin and eosin multi- resolution whole slide images using artificial intelligence. Digital Dentistry Journal. 2026;4(1):100107. doi:10.1016/j.ddj.2026.100107

BibTeX
@article{Sowmya2026,
  author = {Sowmya, K. and Athikary, Dhanya G. and Tiwari, Harsh and Kumar, Harshit and Shivanna, Divya Biligere and Rao, Roopa S. and Astekar, Madhusudan},
  title = {Automating the prognosis of unicystic ameloblastoma on haematoxylin and eosin multi- resolution whole slide images using artificial intelligence},
  journal = {Digital Dentistry Journal},
  year = {2026},
  volume = {4},
  number = {1},
  pages = {100107},
  doi = {10.1016/j.ddj.2026.100107}
}
Kostenlos starten →