EDA-UNet++: EfficientNet-B4 Dual-Attention U-Net++ for Lesion-Wise DR Segmentation

DOI

10.2991/978-94-6239-697-5_22How to use a DOI?

Keywords

Multi-scale lesion detection; UNet-family; EfficientNet-B4; nested skips; dual attention mechanisms; cross-dataset validation

Abstract

For automated screening and clinical decision support, accu-rate segmentation of diabetic retinopathy (DR) lesions is crucial. How-ever, this is still difficult because of lesion variability, low contrast struc-tures, and semantic discrepancies between encoder and decoder represen-tations in convolutional networks. In order to promote accurate retinal lesion delineation, this study presents EDA-UNet++, a dual-attention improved U-Net++ framework. Three essential elements are integrated in the suggested architecture: dense nested skip pathways for progressive semantic alignment, an EfficientNet-B4 encoder for enhanced feature preservation during downsampling, and spatial–channel dual attention refinement to highlight lesion-relevant responses while reducing noise. All baseline models, including U-Net, U-Net++, and Attention U-Net, used the same encoders to guarantee a fair comparison. Experimental assessment on the DDR dataset shows better macro and lesion-wise segmentation performance, with the best validation Dice score among comparison models and significant improvements for exudates and soft exudates. Stable generalization with no loss in performance is confirmed by cross-dataset evaluation on IDRiD. According to computational research, significant gains are made with little extra overhead, confirming their applicability for extensive clinical screening applications.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Akansha Gupta
AU  - Arjun Singh Rawat
AU  - Gunjan Rehani
PY  - 2026
DA  - 2026/06/04
TI  - EDA-UNet++: EfficientNet-B4 Dual-Attention U-Net++ for Lesion-Wise DR Segmentation
BT  - Proceedings of the Conference on Bridging Engineering Disciplines with AI and Machine Learning (BEDAIML 2026)
PB  - Atlantis Press
SP  - 255
EP  - 269
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-697-5_22
DO  - 10.2991/978-94-6239-697-5_22
ID  - Gupta2026
ER  -