Prior-AttUNet for Retinal OCT Fluid Segmentation
Published:Dec 25, 2025 14:37
•1 min read
•ArXiv
Analysis
This paper introduces Prior-AttUNet, a novel deep learning model for segmenting fluid regions in retinal OCT images. The model leverages anatomical priors and attention mechanisms to improve segmentation accuracy, particularly addressing challenges like ambiguous boundaries and device heterogeneity. The high Dice scores across different OCT devices and the low computational cost suggest its potential for clinical application.
Key Takeaways
- •Proposes Prior-AttUNet, a novel model for retinal OCT fluid segmentation.
- •Integrates anatomical priors and attention mechanisms to improve accuracy.
- •Achieves high Dice scores across multiple OCT devices.
- •Demonstrates a balance between segmentation precision and inference efficiency (low computational cost).
Reference
“Prior-AttUNet achieves excellent performance across three OCT imaging devices (Cirrus, Spectralis, and Topcon), with mean Dice similarity coefficients of 93.93%, 95.18%, and 93.47%, respectively.”