EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Automatic classification of Diabetic Retinopathy using Artificial Intelligence models

Milkes Yone Alvarenga, Evandro Catelani Ferraz, Nelson Augusto Oliveira de Aguiar, Ana Beatriz do Nascimento Schlink, Anderson Hideki Suzuki, Bianca Ferreira dos Santos, Pedro Camasmie Kobayashi

Revista DCS · 23.06.2026 · 23 (91) · e5800

DOI: 10.54899/dcs.v23i91.5800 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Revista DCS, online erschienen am 23.06.2026. Verfasst von Milkes Yone Alvarenga und 6 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Diabetic Retinopathy (DR) is a leading cause of adult blindness and, due to its silent progression, requires early detection. This study aimed to develop artificial intelligence models to assist in diagnosing DR, focusing on automated screening and disease stage classification. A balanced dataset was constructed from the APTOS-2019 and BRSET databases. Using the PyTorch library and the ResNet-50 architecture with transfer learning and fine-tuning, two models were developed: a binary screening model and a multiclass staging model. The binary model achieved 94.5% accuracy and an AUC of 0.97, demonstrating its potential as an initial screening tool. The initial multiclass model showed moderate performance (Macro F1-score of 0.59), with difficulties mainly in intermediate stages. A refinement stage was conducted evaluating DenseNet121, EfficientNet-B3, and ConvNeXt-Tiny architectures with regularization techniques such as label smoothing, Dropout, and class weights, resulting in the selection of ConvNeXt-Tiny as the final model, with a Macro F1-score of 0.63. The models were integrated into an API built with FastAPI, with potential for use in telemedicine platforms. The results confirm the binary model's potential as an effective screening tool and identify critical challenges in multiclass classification, outlining directions for future research.

Zitieren

APA

Alvarenga, M. Y., Ferraz, E. C., Aguiar, N. A. O. D., Schlink, A. B. D. N., Suzuki, A. H., Santos, B. F. D., & Kobayashi, P. C. (2026). Automatic classification of Diabetic Retinopathy using Artificial Intelligence models. Revista DCS, 23(91), e5800. https://doi.org/10.54899/dcs.v23i91.5800

Vancouver

Alvarenga MY, Ferraz EC, Aguiar NAOD, Schlink ABDN, Suzuki AH, Santos BFD, et al. Automatic classification of Diabetic Retinopathy using Artificial Intelligence models. Revista DCS. 2026;23(91):e5800. doi:10.54899/dcs.v23i91.5800

BibTeX
@article{Alvarenga2026,
  author = {Alvarenga, Milkes Yone and Ferraz, Evandro Catelani and Aguiar, Nelson Augusto Oliveira de and Schlink, Ana Beatriz do Nascimento and Suzuki, Anderson Hideki and Santos, Bianca Ferreira dos and Kobayashi, Pedro Camasmie},
  title = {Automatic classification of Diabetic Retinopathy using Artificial Intelligence models},
  journal = {Revista DCS},
  year = {2026},
  volume = {23},
  number = {91},
  pages = {e5800},
  doi = {10.54899/dcs.v23i91.5800}
}
Kostenlos starten →