EUVIMEDEUVIMED entdecken
Crossref

Hybrid CNN with depthwise separable convolutions for lightweight diabetic retinopathy detection

Sachin Ahuja, Abhishek Kumar, Pramod Singh Rathore

IET Conference Proceedings · 01.08.2026 · 2025 (63) · 531-536

DOI: 10.1049/icp.2026.2520 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IET Conference Proceedings, erschienen am 01.08.2026. Verfasst von Sachin Ahuja und 2 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

One of the most commonly occurring complications of diabetes is Diabetic Retinopathy (DR) and a significant cause of blindness across the world. The paper provides a relatively small-sized Convolutional Neural Network (CNN) model that consists of Depth-wise Separable Convolutions to have the lightweight design that is efficient and nimble in order to predict the stages of DR with high accuracy based on the retinal fundus images. The model provides an opportunity to reduce the computational cost to the minimum without lowering the level of the classification performance, thus can be deployed in a resource-scarce context. This architecture uses depthwise convolutional blocks as well as pointwise convolutional blocks and global average pooling followed by dense classification layer. It was trained and validated using published datasets and an overall testing accuracy of 93.64 and testing loss of 0.0998 were reported. The training process closely converged with less over fitting and the model was insensitive to generalization with unseen data. Such high performance, in combination with low model complexity, demonstrates the possibility of its scalable clinical deployment. In genetic terms, on the whole, this study provides a new straightforward, precise, and computation-wise result-efficient technique to preventively detect diabetic retinopathy at its early stages with the help of contemporary deep learning algorithms.

Zitieren

APA

Ahuja, S., Kumar, A., & Rathore, P. S. (2026). Hybrid CNN with depthwise separable convolutions for lightweight diabetic retinopathy detection. IET Conference Proceedings, 2025(63), 531-536. https://doi.org/10.1049/icp.2026.2520

Vancouver

Ahuja S, Kumar A, Rathore PS. Hybrid CNN with depthwise separable convolutions for lightweight diabetic retinopathy detection. IET Conference Proceedings. 2026;2025(63):531-536. doi:10.1049/icp.2026.2520

BibTeX
@article{Ahuja2026,
  author = {Ahuja, Sachin and Kumar, Abhishek and Rathore, Pramod Singh},
  title = {Hybrid CNN with depthwise separable convolutions for lightweight diabetic retinopathy detection},
  journal = {IET Conference Proceedings},
  year = {2026},
  volume = {2025},
  number = {63},
  pages = {531--536},
  doi = {10.1049/icp.2026.2520}
}
Kostenlos starten →