Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Diabetic Retinopathy Lesion Segmentation through Attention Mechanisms

Aruna Jithesh, Chinmayi Karumuri, Venkata Kiran Reddy Kotha, Meghana Doddapuneni, Taehee Jeong

TL;DR

This paper addresses the need for lesion-level evidence in automated diabetic retinopathy screening by developing Attention-DeepLab, which integrates Convolutional Block Attention Module (CBAM) into DeepLab-V3+ to improve segmentation of four DR-related lesions. The approach employs a dual-path attention mechanism to enhance both high-level semantic and low-level spatial features, yielding notable gains in mean average precision and intersection-over-union, particularly for microaneurysms. On the DDR dataset, the method achieves a 10.5% absolute increase in mAP and a substantial 272% gain in MA detection, underscoring the clinical value of precise lesion localization for early DR intervention. The work demonstrates the potential of attention-based segmentation to provide lesion maps that support ophthalmologists and enable scalable, automated DR screening, with clear directions for external validation and practical deployment.

Abstract

Diabetic Retinopathy (DR) is an eye disease which arises due to diabetes mellitus. It might cause vision loss and blindness. To prevent irreversible vision loss, early detection through systematic screening is crucial. Although researchers have developed numerous automated deep learning-based algorithms for DR screening, their clinical applicability remains limited, particularly in lesion segmentation. Our method provides pixel-level annotations for lesions, which practically supports Ophthalmologist to screen DR from fundus images. In this work, we segmented four types of DR-related lesions: microaneurysms, soft exudates, hard exudates, and hemorrhages on 757 images from DDR dataset. To enhance lesion segmentation, an attention mechanism was integrated with DeepLab-V3+. Compared to the baseline model, the Attention-DeepLab model increases mean average precision (mAP) from 0.3010 to 0.3326 and the mean Intersection over Union (IoU) from 0.1791 to 0.1928. The model also increased microaneurysm detection from 0.0205 to 0.0763, a clinically significant improvement. The detection of microaneurysms is the earliest visible symptom of DR.

Diabetic Retinopathy Lesion Segmentation through Attention Mechanisms

TL;DR

This paper addresses the need for lesion-level evidence in automated diabetic retinopathy screening by developing Attention-DeepLab, which integrates Convolutional Block Attention Module (CBAM) into DeepLab-V3+ to improve segmentation of four DR-related lesions. The approach employs a dual-path attention mechanism to enhance both high-level semantic and low-level spatial features, yielding notable gains in mean average precision and intersection-over-union, particularly for microaneurysms. On the DDR dataset, the method achieves a 10.5% absolute increase in mAP and a substantial 272% gain in MA detection, underscoring the clinical value of precise lesion localization for early DR intervention. The work demonstrates the potential of attention-based segmentation to provide lesion maps that support ophthalmologists and enable scalable, automated DR screening, with clear directions for external validation and practical deployment.

Abstract

Diabetic Retinopathy (DR) is an eye disease which arises due to diabetes mellitus. It might cause vision loss and blindness. To prevent irreversible vision loss, early detection through systematic screening is crucial. Although researchers have developed numerous automated deep learning-based algorithms for DR screening, their clinical applicability remains limited, particularly in lesion segmentation. Our method provides pixel-level annotations for lesions, which practically supports Ophthalmologist to screen DR from fundus images. In this work, we segmented four types of DR-related lesions: microaneurysms, soft exudates, hard exudates, and hemorrhages on 757 images from DDR dataset. To enhance lesion segmentation, an attention mechanism was integrated with DeepLab-V3+. Compared to the baseline model, the Attention-DeepLab model increases mean average precision (mAP) from 0.3010 to 0.3326 and the mean Intersection over Union (IoU) from 0.1791 to 0.1928. The model also increased microaneurysm detection from 0.0205 to 0.0763, a clinically significant improvement. The detection of microaneurysms is the earliest visible symptom of DR.
Paper Structure (16 sections, 11 equations, 6 figures, 1 table)

This paper contains 16 sections, 11 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. DeepLab-V3+
  4. Convolutional Block Attention Module
  5. U-Net and Gated U-Net
  6. Methodology
  7. Dataset
  8. Data Preprocessing
  9. DeepLab-V3+ with Attention Mechanisms
  10. Loss Function
  11. Implementation Details
  12. Evaluation Metrics
  13. Experimental Results
  14. Discussion
  15. Future Work
  16. ...and 1 more sections

Figures (6)

  • Figure 1: (a) Normal retina (b) Diabetic Retina images from DDR dataset
  • Figure 2: Neural architecture of (a) DeepLab-V3 (b) DeepLab-V3+
  • Figure 3: Convolutional Block Attention Module. B: batch size, C: channel of feature map, H: height of feature map, W: width of feature map, $r$: reduction ratio.
  • Figure 4: Neural architecture of Attention-DeepLab
  • Figure 5: Segmentation example 1: (a) Input image, (b) Ground Truth, and (c) Prediction
  • ...and 1 more figures