https://doi.org/10.1140/epjs/s11734-024-01334-9
Regular Article
Revolutionizing diabetic retinopathy detection using DB-SCA-UNet with Drop Block-Based Attention Model in deep learning for precise analysis of color retinal images
School of Computer Science and Engineering, VIT-AP University, 522237, Vijayawada, India
Received:
22
June
2024
Accepted:
5
September
2024
Published online:
20
September
2024
Diabetic retinopathy (DR) is a widespread retinal illness that manifests as swollen retinal vessels and elevated blood sugar levels. Detecting and screening for DR early on can increase the likelihood of positive outcomes. Manual screening for retinal detachment is a laborious and time-consuming process. In recent years, deep learning (DL) has significantly impacted retinal vascular segmentation. Accurately segmenting retinal vessels is essential for computer-assisted diagnosis of eye disorders, including diabetes and glaucoma. U-Net architectures are predominantly utilized in biomedical image segmentation to automate the identification and detection of target areas or sub-regions. U-Net-based algorithms have been frequently applied in retinal vascular segmentation tasks. However, significant challenges remain, such as the loss of microvasculature features at vessel ends and the interference caused by lesion-produced hard exudates. In this paper, we propose an adaptive system named Drop Block-based Spatial-Channel Attention U-Net (DB-SCA-UNet), which uses a symmetrical structure to enhance locally relevant features while suppressing irrelevant features at the spatial and channel levels. In addition, to mitigate overfitting, the proposed method replaces the original U-Net convolutional blocks with channel dropout convolutional blocks. The effectiveness of this novel architecture was assessed using three public datasets—DRIVE, STARE, and CHASE_DB1—as well as one custom dataset, LDDR. Experimental results demonstrate that our DB-SCA-UNet can accurately and efficiently segment retinal vessels. The graphical representation of the findings is represented in Fig. 1.
Copyright comment Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
© The Author(s), under exclusive licence to EDP Sciences, Springer-Verlag GmbH Germany, part of Springer Nature 2024. Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
