Related Experiment Video
Updated: Aug 9, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
VEA-net: vascular enhancement attention with dual-backbone multi-task learning for comprehensive ROP management
Yuan Chen1, Jiajun Wan2, Tian Zhang1
1Department of Pediatrics, Shandong Provincial Hospital Affiliated to Shandong First Medical University, Jinan, Shandong, China.
Frontiers in Cell and Developmental Biology
|August 8, 2026
Summary
This study introduces VEA-Net, a deep learning model for Retinopathy of Prematurity (ROP) management. VEA-Net enhances vascular features and integrates multiple tasks for improved detection, classification, and treatment decisions in ROP.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Retinopathy of Prematurity (ROP) is a leading cause of childhood blindness.
- Current clinical ROP management involves complex, correlated tasks: Plus disease detection, stage classification, and treatment decisions.
- Deep learning (DL) approaches face challenges in modeling vascular morphology, integrating multi-task learning, and ensuring robustness across diverse datasets.
Purpose of the Study:
- To develop a unified deep learning framework for automated Retinopathy of Prematurity (ROP) management.
- To explicitly model and enhance retinal vascular features for improved Plus disease assessment.
- To jointly optimize ROP detection, classification, and treatment decision-making while ensuring model robustness across heterogeneous datasets.
Main Methods:
- Proposed VEA-Net, a dual-backbone multi-task learning framework.
- Introduced a Vascular Enhancement Attention (VEA) module for explicit vascular feature modeling.
- Implemented hierarchical multi-task learning and a dual-domain adaptation strategy (DANN, MMD) for robustness.
Main Results:
- Validated VEA-Net on three public ROP datasets (8,636 images).
- Achieved high AUROC scores: 91.76% for Plus detection, 88.26% for stage classification.
- Demonstrated robust performance across heterogeneous datasets, indicating potential for improved ROP management efficiency.
Conclusions:
- VEA-Net offers a unified framework for ROP management, integrating vascular enhancement, multi-task learning, and domain adaptation.
- The model supports multiple clinical tasks within a single architecture.
- VEA-Net shows potential for enhancing the efficiency and robustness of automated ROP management.
