Concat U-Net Classifier with Attention Enhanced DL framework for Diabetic Retinopathy Classification
Procedia Computer Science · 2026 · 283 · 4688-4696
DOI: 10.1016/j.procs.2026.06.524 ↗Kurz gefasst
Fachartikel in der Zeitschrift Procedia Computer Science, 2026. Verfasst von Sandhya Kanoujia und 1 weiteren Autor:innen. Die Arbeit nennt 26 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Diabetic Retinopathy (DR) is a microvascular condition of diabetes that leads to irreversible impaired vision if neglected during the initial stages. Thus, it is crucial to have a computerized system that can study the features of a retinal fundus image, find the disease and suggest the treatment. Deep learning algorithms have made notable and significant contributions to this area. Although these methods achieve high accuracy, they fail to preserve discriminative lesion features such as microaneurysms and hemorrhages. To address this limitation, a Concat U-Net classifier with attention-augmented skip connections and sharpened feature preservation is proposed in this study. The Convolutional Block Attention Module (CBAM) emphasizes clinically relevant regions, though sharpen filters mitigate feature blurring across encoder–decoder pathways. The highlight is integrating the potential of saliency-preserving skip connections in U-Net–style architectures for robust retinal disease diagnosis. The architecture concatenates intermediate convolutional features to preserve fine-grained lesion details, ensuring robustness across multiple severity levels. The simulation results on two widely used datasets, one is the APTOS 2019 blindness detection dataset and the other is the IDRiD dataset, validate that the projected model attains 0.89 to 0.93 in terms of AUC and 0.85 Kappa score, outperforming standard Convolutional Neural Networks (CNN) and Transformer-based DR classifiers.
Zitieren
APA
Kanoujia, S., & Pitchai, K. (2026). Concat U-Net Classifier with Attention Enhanced DL framework for Diabetic Retinopathy Classification. Procedia Computer Science, 283, 4688-4696. https://doi.org/10.1016/j.procs.2026.06.524
Vancouver
Kanoujia S, Pitchai K. Concat U-Net Classifier with Attention Enhanced DL framework for Diabetic Retinopathy Classification. Procedia Computer Science. 2026;283:4688-4696. doi:10.1016/j.procs.2026.06.524
BibTeX
@article{Kanoujia2026,
author = {Kanoujia, Sandhya and Pitchai, Karuppanan},
title = {Concat U-Net Classifier with Attention Enhanced DL framework for Diabetic Retinopathy Classification},
journal = {Procedia Computer Science},
year = {2026},
volume = {283},
pages = {4688--4696},
doi = {10.1016/j.procs.2026.06.524}
}
EUVIMED