SDC-Net - Enhancing Multiclass Diabetic Retinopathy Segmentation by Shallow-Deep Collaboration Networks With Wavelet-Guided Attention Mechanism

Sep 28, 2026

Diabetic retinopathy (DR) is a major cause of vision loss, and accurate segmentation of fundus lesions is critical for screening, grading, and monitoring. Segmentation is difficult because different lesion types vary in shape, size, and appearance, yet four key lesion classes must be identified simultaneously. SDC-Net presents an end-to-end shallow–deep collaboration framework using wavelet-guided attention and a super-resolution auxiliary task, evaluated on IDRiD, DDR, and FGADR.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages16

Reading Time40 min

AccessFree to download

Purpose

Methods

Proposed SDC-Net framework

Wavelet-guided attention and super-resolution auxiliary task

Results

Conclusions

Translational Relevance

Introduction

Methods SDC-Net: Enhancing Multiclass Diabetic Retinopathy Segmentation by Shallow–Deep Collaboration Networks With Wavelet-Guided Attention Mechanism Jiaoli Liu 1, Lin Zhang 1, Zhen Liu 1, Yonggang Xu 1, Xiaojuan Wang2,3, and Hui Liu 1 1 School of Information and Control Engineering, China University of Mining and Technology, Xuzhou, China 2 The Department of Ophthalmology, Shanghai First People’s Hospital, Affiliated to Shanghai Jiao Tong University, Shanghai, China 3 Ophthalmology and Optometry Center, Xuzhou Municipal First People’s Hospital, Xuzhou, China Correspondence: Hui Liu, School of Information and Control Engineering, China University of Mining and Technology, Xuzhou 221116, China. e-mail: [hui.liu@cumt.edu.cn](hui.liu@cumt.edu.cn) Received: July 23, 2025 Accepted: November 14, 2025 Published: December 29, 2025 Keywords: diabetic retinopathy; wavelet transform; lesion segmentation; multiscale; super-resolution Citation: Liu J, Zhang L, Liu Z, Xu Y, Wang X, Liu H. SDC-Net: Enhancing multiclass diabetic retinopathy segmentation by shallow–deep collaboration networks with wavelet-guided attention mechanism. Transl Vis Sci Technol. 2025;14(12):28, [https://doi.org/10.1167/tvst.14.12.28](https://doi.org/10.1167/tvst.14.12.28) Purpose: Diabetic retinopathy (DR) is a leading cause of blindness. Fundus lesions are key clinical signs of DR, and their accurate segmentation is crucial for screening, grading, and monitoring the disease. However, segmenting different lesions simultaneously is challenging owing to their varying shapes, sizes, and appearances. This work aimed to segment four types of DR lesions simultaneously. Methods: We propose a shallow–deep collaboration network with a wavelet-guided attention mechanism for simultaneous segmentation of four DR-related lesions. Our end-to-end framework integrates shallow and deep networks to enhance multiscale feature extraction. The deep network uses a wavelet-based attention mechanism to fuse multiscale context representations. Additionally, a super-resolution auxiliary task is introduced to improve training accuracy. Results: Extensive experiments are conducted on the IDRiD, DDR, and FGADR public datasets for evaluation. The average Dice scores of SDC-Net on the three datasets are 60.27, 39.98, and 42.57, with average intersection over union scores of 44.53, 25.71, and 28.58, and average area under the receiver operating characteristic curve values of 68.75, 57.37, and 64 .21. Conclusions: The dual-branch design in the proposed framework enables better capture of multiscale features and improves segmentation. The dual wavelet attention module of the deep branch can enhance the extraction of detailed lesion features, and the introduced super-resolution task further improves the accuracy and robustness. Translational Relevance: The SDC-Net framework has the potential to enhance the clinical diagnosis of DR by offering more precise segmentation of multiple types of fundus lesions, thereby aiding in early detection and management of the disease. Introduction Diabetic retinopathy (DR) stands as a prevalent complication of diabetes and a predominant cause of vision loss worldwide.1 , 2 Consequently, there is a pressing need for its early detection and diagnosis. In clinical practice, DR severity is classified into five stages (range, 0–4) following international protocols,3 where stages 1 to 3 (nonproliferative DR) are diagnostically reliant on the identification of four cardinal fundus lesions: microaneurysms (MA), hemorrhages (HE), soft exudates (SE), and hard exudates (EX) . As shown in Figure 1, two examples are provided to contrast a normal image with one with multiclass lesions. Crucially, early detection of these lesions is pivotal for timely intervention, potentially halting progression to vision-threatening proliferative DR (stage 4) . Owing to its convenience and noninvasiveness, color photography has become a widely used fundus

FAQs

What problem does SDC-Net address in diabetic retinopathy imaging?

It targets simultaneous segmentation of four DR lesion types from fundus images, where lesion heterogeneity makes multi-class segmentation challenging for existing approaches.

How does SDC-Net improve multiscale feature learning?

It combines shallow and deep networks for stronger multiscale feature extraction, and the deep branch uses wavelet-based attention to fuse multiscale contextual representations.

What training strategy supports SDC-Net’s accuracy and robustness?

A super-resolution auxiliary task is introduced alongside the segmentation framework to improve training accuracy and generalization.

Sep 28, 2026

Uploaded by A glass of water