Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Convolutional Prompting for Broad-Domain Retinal Vessel Segmentation

Qijie Wei, Weihong Yu, Xirong Li

TL;DR

This work tackles broad-domain retinal vessel segmentation (BD-RVS) by introducing Dual Convolutional Prompting (DCP), a plug-in module that injects domain-specific prompts into an existing R2AU-Net segmentation network without altering its architecture. DCP prompts along position and channel dimensions within localized $8\times8$ feature patches, using self-attention with sequence-length reduction to extract domain-specific vessel cues across CFP, SLO, UWF, OCTA, and FFA. On a broad-domain dataset assembled from five public sources, DCP outperforms eight baselines, achieving higher mean AP (0.7037) with fewer parameters, and ablations confirm the necessity of both prompt types and locality. The approach demonstrates that domain-aware, localized prompting can unify cross-domain medical image segmentation tasks and suggests broad applicability to multi-domain retinal analysis.

Abstract

Previous research on retinal vessel segmentation is targeted at a specific image domain, mostly color fundus photography (CFP). In this paper we make a brave attempt to attack a more challenging task of broad-domain retinal vessel segmentation (BD-RVS), which is to develop a unified model applicable to varied domains including CFP, SLO, UWF, OCTA and FFA. To that end, we propose Dual Convoltuional Prompting (DCP) that learns to extract domain-specific features by localized prompting along both position and channel dimensions. DCP is designed as a plug-in module that can effectively turn a R2AU-Net based vessel segmentation network to a unified model, yet without the need of modifying its network structure. For evaluation we build a broad-domain set using five public domain-specific datasets including ROSSA, FIVES, IOSTAR, PRIME-FP20 and VAMPIRE. In order to benchmark BD-RVS on the broad-domain dataset, we re-purpose a number of existing methods originally developed in other contexts, producing eight baseline methods in total. Extensive experiments show the the proposed method compares favorably against the baselines for BD-RVS.

Convolutional Prompting for Broad-Domain Retinal Vessel Segmentation

TL;DR

This work tackles broad-domain retinal vessel segmentation (BD-RVS) by introducing Dual Convolutional Prompting (DCP), a plug-in module that injects domain-specific prompts into an existing R2AU-Net segmentation network without altering its architecture. DCP prompts along position and channel dimensions within localized feature patches, using self-attention with sequence-length reduction to extract domain-specific vessel cues across CFP, SLO, UWF, OCTA, and FFA. On a broad-domain dataset assembled from five public sources, DCP outperforms eight baselines, achieving higher mean AP (0.7037) with fewer parameters, and ablations confirm the necessity of both prompt types and locality. The approach demonstrates that domain-aware, localized prompting can unify cross-domain medical image segmentation tasks and suggests broad applicability to multi-domain retinal analysis.

Abstract

Previous research on retinal vessel segmentation is targeted at a specific image domain, mostly color fundus photography (CFP). In this paper we make a brave attempt to attack a more challenging task of broad-domain retinal vessel segmentation (BD-RVS), which is to develop a unified model applicable to varied domains including CFP, SLO, UWF, OCTA and FFA. To that end, we propose Dual Convoltuional Prompting (DCP) that learns to extract domain-specific features by localized prompting along both position and channel dimensions. DCP is designed as a plug-in module that can effectively turn a R2AU-Net based vessel segmentation network to a unified model, yet without the need of modifying its network structure. For evaluation we build a broad-domain set using five public domain-specific datasets including ROSSA, FIVES, IOSTAR, PRIME-FP20 and VAMPIRE. In order to benchmark BD-RVS on the broad-domain dataset, we re-purpose a number of existing methods originally developed in other contexts, producing eight baseline methods in total. Extensive experiments show the the proposed method compares favorably against the baselines for BD-RVS.

Paper Structure

This paper contains 9 sections, 1 equation, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. The Segmentation Network
  4. The Dual Convolutional Prompting Module
  5. Experiments
  6. Experimental Setup
  7. Comparison with SOTA
  8. Ablation study
  9. Conclusions

Figures (5)

  • Figure 1: Two paradigms for retinal vessel segmentation (RVS): (a) narrow-domain and (b) broad-domain. This paper aims for the latter.
  • Figure 2: Our DCP method for broad-domain RVS
  • Figure 3: Proposed dual convolutional prompting (DCP) module. Its input is the output feature map of a specific encoder of R2AU-Net. Based on the domain identity of the input image, DCP takes two domain-specific prompt tensors, which interact with the feature maps along the position and channel dimensions, respectively. For localized prompting, the feature map is partitioned into smaller (orange) patches. Once trained, the prompts are fixed. Best viewed in color.
  • Figure 4: Visualization of the input ($F_1$) and output ($F^t_1$) of DCP. For all the five modalities, vessel-related patterns are noticeably enhanced.
  • Figure 5: Qualitative results. The efficacy of DCP is primarily manifested in segmenting capillaries. Best viewed digitally.