Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Global and Patch-wise Contrastive Loss for Accurate Automated Exudate Detection

Wei Tang, Kangning Cui, Raymond H. Chan

TL;DR

This work tackles hard exudate segmentation in retinal fundus images by introducing a supervised contrastive framework that combines patch-wise density contrasting and discriminative edge inspection. The approach jointly optimizes $L_{total} = L_{sup} + \alpha\,(L_{pd} + L_{de})$, enabling density-aware and boundary-aware representations to improve segmentation, particularly for tiny and boundary-ambiguous lesions. On the IDRiD dataset, it achieves state-of-the-art IoU, F1, and recall while remaining robust across multiple backbones, demonstrating practical potential for computer-assisted DR screening. The combination of density-aware patch contrasts and edge-focused features provides a scalable, generalizable strategy for challenging medical segmentation tasks with irregular lesion distributions and boundary uncertainty.

Abstract

Diabetic retinopathy (DR) is a leading global cause of blindness. Early detection of hard exudates plays a crucial role in identifying DR, which aids in treating diabetes and preventing vision loss. However, the unique characteristics of hard exudates, ranging from their inconsistent shapes to indistinct boundaries, pose significant challenges to existing segmentation techniques. To address these issues, we present a novel supervised contrastive learning framework to optimize hard exudate segmentation. Specifically, we introduce a patch-wise density contrasting scheme to distinguish between areas with varying lesion concentrations, and therefore improve the model's proficiency in segmenting small lesions. To handle the ambiguous boundaries, we develop a discriminative edge inspection module to dynamically analyze the pixels that lie around the boundaries and accurately delineate the exudates. Upon evaluation using the IDRiD dataset and comparison with state-of-the-art frameworks, our method exhibits its effectiveness and shows potential for computer-assisted hard exudate detection. The code to replicate experiments is available at github.com/wetang7/HECL/.

A Global and Patch-wise Contrastive Loss for Accurate Automated Exudate Detection

TL;DR

This work tackles hard exudate segmentation in retinal fundus images by introducing a supervised contrastive framework that combines patch-wise density contrasting and discriminative edge inspection. The approach jointly optimizes , enabling density-aware and boundary-aware representations to improve segmentation, particularly for tiny and boundary-ambiguous lesions. On the IDRiD dataset, it achieves state-of-the-art IoU, F1, and recall while remaining robust across multiple backbones, demonstrating practical potential for computer-assisted DR screening. The combination of density-aware patch contrasts and edge-focused features provides a scalable, generalizable strategy for challenging medical segmentation tasks with irregular lesion distributions and boundary uncertainty.

Abstract

Diabetic retinopathy (DR) is a leading global cause of blindness. Early detection of hard exudates plays a crucial role in identifying DR, which aids in treating diabetes and preventing vision loss. However, the unique characteristics of hard exudates, ranging from their inconsistent shapes to indistinct boundaries, pose significant challenges to existing segmentation techniques. To address these issues, we present a novel supervised contrastive learning framework to optimize hard exudate segmentation. Specifically, we introduce a patch-wise density contrasting scheme to distinguish between areas with varying lesion concentrations, and therefore improve the model's proficiency in segmenting small lesions. To handle the ambiguous boundaries, we develop a discriminative edge inspection module to dynamically analyze the pixels that lie around the boundaries and accurately delineate the exudates. Upon evaluation using the IDRiD dataset and comparison with state-of-the-art frameworks, our method exhibits its effectiveness and shows potential for computer-assisted hard exudate detection. The code to replicate experiments is available at github.com/wetang7/HECL/.
Paper Structure (14 sections, 6 equations, 3 figures, 2 tables)

This paper contains 14 sections, 6 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Medical Image Segmentation
  4. Hard Exudate Detection
  5. Contrastive Learning
  6. Methodology
  7. Patch-wise Density Contrasting
  8. Discriminative Edge Inspection
  9. Experiments
  10. Experimental Setup
  11. Results and Discussion
  12. Conclusion and Future Works
  13. Compliance with Ethical Standards
  14. Acknowledgement

Figures (3)

  • Figure 1: Illustration of hard exudates in a fundus image. The red pixels in the right image locate the hard exudate lesions.
  • Figure 2: An overview of the proposed framework. The network is jointly trained by $\mathcal{L}_{sup}$, $\mathcal{L}_{pd}$, and $\mathcal{L}_{pe}$ in order to learn both "density-aware" and "boundary-aware" knowledge.
  • Figure 3: Comparison of network segmentations with (w/ CL) and without (w/o CL) the proposed framework. Top row: input images and GT masks. Following rows: segmentation outputs, with yellow boxes emphasizing optic disc and small lesion areas.