Context guided network improves retinal vessel segmentation accuracy
CGDD-Net: Context-Guided Dynamic Detail Modeling for Retinal Vessel Segmentation
Computer Vision and Pattern Recognition
Summary
Segmenting tiny blood vessels in retinal images is tricky because it requires understanding both the overall shape and fine details. The authors designed a new computer model called CGDD-Net that uses smart attention mechanisms to capture vessel patterns at different sizes and adaptively focuses on important regions. Their method combines and compresses intermediate features before decoding, which helps produce clearer vessel maps with fewer parameters. Tests on several standard datasets show this approach improves accuracy compared to previous methods.
What this means in practice
- •For medical imaging teams: Enhance automated retinal vessel segmentation in diagnostic imaging systems to improve detection of eye diseases.
- •For healthcare ai developers: Incorporate a compact and accurate vessel segmentation model into retinal analysis software for telemedicine platforms.$Commercial implications: Enables development of commercial retinal diagnostic tools with higher accuracy and smaller model sizes suitable for deployment.
Authors
Xincheng Li, Xiaoqi Sheng, Xinyu Zhang
Abstract
Accurate retinal vessel segmentation requires features that capture vascular geometry while preserving information for fine-scale reconstruction. We propose CGDD-Net, a context-guided dynamic detail modeling network that connects adaptive feature extraction to a shared decoder pathway. Context-Guided Scale-Adaptive Deformable Encoding (CSDE) combines fixed-grid convolution with deformable local attention to capture vascular patterns at multiple spatial extents. Spatially Adaptive Multi-Kernel Gating (SAMG) selects receptive-field responses at each location. Dynamic Cross-Scale Detail Fusion (DCDF) aligns the gated intermediate features and compresses them into an eight-channel representation, which is reused at three decoder resolutions together with selected encoder skips. This design consolidates intermediate information before decoding instead of transferring each middle-stage feature through a separate direct skip. On DRIVE, CHASE\_DB1, STARE, and HRF, CGDD-Net achieves AUC values of 0.9824, 0.9938, 0.9895, and 0.9874, with F1 scores of 0.8323, 0.8102, 0.8510, and 0.8157, respectively. The complete model contains 1.96 million trainable parameters. In cumulative ablations, the full model improves F1 over the internal baseline by 2.48, 0.61, and 3.49 percentage points on DRIVE, CHASE\_DB1, and STARE. Twelve directed cross-dataset experiments further characterize transfer without target-domain adaptation. The results support shared intermediate detail delivery as an effective, parameter-compact architecture for retinal vessel segmentation. Code is available at \url{https://github.com/lixincheng-xcl/CGDD-Net}.