EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Synergistic Stacked Ensemble Deep Learning Model for Predicting Diabetic Retinopathy Severity

Deva Kumar Salluri, Maanas Sai Surya Chandra Atthuluri, Venkatramaphanikumar Sistla, Venkata Krishna Kishore Kolli

IEEE Access · 2025 · 13 · 202454-202466

DOI: 10.1109/access.2025.3637311 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2025. Verfasst von Deva Kumar Salluri und 3 weiteren Autor:innen. Die Arbeit wurde bisher 6-mal zitiert (Crossref) und nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Diabetic retinopathy is a leading cause of vision loss worldwide. Early and accurate diagnosis can prevent vision loss, and automatic grading is particularly important to assist physicians in making timely decisions. This work proposes an efficient stacked ensemble framework for the DR severity classification task. Unlike traditional ensembles, our model leverages out-of-fold predictions and collaborative meta-learning to improve ordinal consistency without leaking information during training. We incorporate the best features of three top deep learning architectures: Swin-B Transformer, EfficientNetV2-L, and ConvNeXt-L. A meta-learner maximizes their collective capacity for prediction. Extensive experiments on two widely available datasets, the APTOS 2019 Blindness Detection dataset and a larger Kaggle dataset, were conducted to evaluate our approach. We investigated the generalization capability of our ensemble with both 3-fold and 5-fold cross-validation. With the highest classification accuracy of 97.9% for the APTOS dataset and 98.2% for the Kaggle dataset, the results clearly indicate that our stacked model outperforms all individual networks. Furthermore, for both datasets, we consistently achieved a high QWK value of 0.97. This reflects strong agreement between our predictions and professional clinical evaluations. It also demonstrates the goodness of our model dealing with ordinal data. These findings emphasize that there is great potential in Stacked Ensembling to enhance the accuracy and robustness of automated diabetic retinopathy grading systems, particularly when combined with diverse high performance base models.

Zitieren

APA

Salluri, D. K., Atthuluri, M. S. S. C., Sistla, V., & Kolli, V. K. K. (2025). A Synergistic Stacked Ensemble Deep Learning Model for Predicting Diabetic Retinopathy Severity. IEEE Access, 13, 202454-202466. https://doi.org/10.1109/access.2025.3637311

Vancouver

Salluri DK, Atthuluri MSSC, Sistla V, Kolli VKK. A Synergistic Stacked Ensemble Deep Learning Model for Predicting Diabetic Retinopathy Severity. IEEE Access. 2025;13:202454-202466. doi:10.1109/access.2025.3637311

BibTeX
@article{Salluri2025,
  author = {Salluri, Deva Kumar and Atthuluri, Maanas Sai Surya Chandra and Sistla, Venkatramaphanikumar and Kolli, Venkata Krishna Kishore},
  title = {A Synergistic Stacked Ensemble Deep Learning Model for Predicting Diabetic Retinopathy Severity},
  journal = {IEEE Access},
  year = {2025},
  volume = {13},
  pages = {202454--202466},
  doi = {10.1109/access.2025.3637311}
}
Kostenlos starten →