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

Explainable deep learning self‐supervised vision transformer for fundus‐based myopia classification

Eirini Maliagkani iD, Christos Tsoutsas, Nikolaos Papageorgiou, Ioannis D Apostolopoulos iD, Nikolaos Papandrianos, Ilias Georgalas, Elpiniki Papageorgiou

Acta Ophthalmologica · 27.07.2026

DOI: 10.1111/aos.70205 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Acta Ophthalmologica, online erschienen am 27.07.2026. Verfasst von Eirini Maliagkani und 6 weiteren Autor:innen. Beteiligte Einrichtung: 1st Department of Ophthalmology, General Hospital of Athens "G. Gennimatas", National and Kapodistrian University of Athens, Athens, Greece. Die Arbeit nennt 44 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Purpose

To develop and validate a self‐supervised vision transformer (ViT) for automated three‐class classification of Normal, High Myopia (HM), and Pathologic Myopia (PM) from colour fundus photographs, with anatomically faithful interpretability.

Methods

A DINOv2 self‐supervised ViT was fine‐tuned using a two‐stage transfer‐learning protocol with class‐balanced sampling. Model robustness was assessed using 10‐fold stratified cross‐validation, and an ensemble of the best fold‐specific checkpoints was evaluated once on a strictly held‐out independent test set. Comparative performance was assessed against ResNet‐50, VGG‐16, and EfficientNet‐B3 using paired, two‐sided Student's t ‐tests on fold‐level macro–F1 scores. Model explainability was examined using a Dual‐Rollout Attention approach that fuses early‐layer spatial detail with late‐layer semantic information.

Results

On the independent test set, the ensemble achieved an overall accuracy of 97.03% (95% CI 96.75–97.32%) and a macro F1‐score of 97.08% (95% CI 96.80–97.36%). The DINOv2‐based framework outperformed ResNet‐50, VGG‐16, and EfficientNet‐B3 architectures (all p < 0.001). Dual‐Rollout Attention highlighted clinically relevant regions including optic disc morphology, peripapillary atrophy, and areas consistent with chorioretinal thinning.

Conclusion

A self‐supervised ViT combined with transformer‐specific interpretability enables accurate, robust, and transparent automated classification of high and pathologic myopia from fundus photographs. This approach may support standardized assessment and scalable screening, pending external and prospective validation.

Schlagwörter

Stichwörter der Autor:innen

deep learningexplainable artificial intelligencefundus photographyhigh myopiapathologic myopiavision transformer

Autor:innen und Institutionen

  • 1st Department of Ophthalmology, General Hospital of Athens "G. Gennimatas", National and Kapodistrian University of Athens, Athens, Greece.
  • ACTA Lab, Department of Energy Systems, University of Thessaly, Larisa, Greece.

ORCID-Profile: Eirini Maliagkani, Ioannis D Apostolopoulos

Zitieren

APA

Maliagkani, E., Tsoutsas, C., Papageorgiou, N., Apostolopoulos, I. D., Papandrianos, N., Georgalas, I., & Papageorgiou, E. (2026). Explainable deep learning self‐supervised vision transformer for fundus‐based myopia classification. Acta Ophthalmologica. https://doi.org/10.1111/aos.70205

Vancouver

Maliagkani E, Tsoutsas C, Papageorgiou N, Apostolopoulos ID, Papandrianos N, Georgalas I, et al. Explainable deep learning self‐supervised vision transformer for fundus‐based myopia classification. Acta Ophthalmologica. 2026. doi:10.1111/aos.70205

BibTeX
@article{Maliagkani2026,
  author = {Maliagkani, Eirini and Tsoutsas, Christos and Papageorgiou, Nikolaos and Apostolopoulos, Ioannis D and Papandrianos, Nikolaos and Georgalas, Ilias and Papageorgiou, Elpiniki},
  title = {Explainable deep learning self‐supervised vision transformer for fundus‐based myopia classification},
  journal = {Acta Ophthalmologica},
  year = {2026},
  doi = {10.1111/aos.70205}
}
Kostenlos starten →