Related Experiment Video
Updated: Jul 12, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
One-shot Retinal Artery and Vein Segmentation via Cross-modality Pretraining
Danli Shi1,2,3, Shuang He3, Jiancheng Yang4
1Centre for Eye and Vision Research (CEVR), Hong Kong SAR, China.
Ophthalmology Science
|October 23, 2023
Summary
This study introduces a novel one-shot method for retinal artery and vein segmentation using cross-modality pretraining. The approach achieves high accuracy with minimal data, offering an efficient solution for retinal vessel analysis.
Area of Science:
- Ophthalmology
- Medical Imaging
- Computer Vision
Background:
- Accurate segmentation of retinal arteries and veins is crucial for diagnosing various eye conditions.
- Traditional methods often require large, annotated datasets, which are time-consuming and expensive to acquire.
Purpose of the Study:
- To develop and evaluate a one-shot learning approach for retinal artery-vein segmentation.
- To leverage cross-modality artery-vein (AV) soft-label pretraining to improve segmentation performance with limited data.
Main Methods:
- Utilized a large dataset of color fundus photography (CFP) and arterial-venous fundus fluorescein angiography (FFA) pairs for pretraining.
- Generated AV soft labels automatically from FFA images based on intensity differences.
- Trained a generative adversarial network (GAN) for AV soft segmentation using CFP images.
- Finetuned the pretrained model using only one image per dataset for one-shot segmentation.
Main Results:
- The one-shot approach achieved performance comparable to full-data training, with AUCs from 0.901 to 0.971 and accuracy from 0.959 to 0.980.
- One-shot finetuning significantly improved segmentation performance compared to no finetuning.
- Segmentation results demonstrated high reliability, with low standard deviations across different finetuning images.
Conclusions:
- This study presents the first one-shot method for retinal artery and vein segmentation.
- The proposed automatic soft-labeling and pretraining strategy is time-saving, efficient, and shows potential for widespread application in retinal imaging analysis.
