Related Experiment Video
Updated: Jun 12, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
RetVes segmentation: A pseudo-labeling and feature knowledge distillation optimization technique for retinal vessel
Favour Ekong1, Yongbin Yu1, Rutherford Agbeshi Patamia1
1School of Information and Software Engineering, University of Electronic Science and Technology of China, Chengdu, 610054, Sichuan, China.
Computers in Biology and Medicine
|September 19, 2024
Summary
This study introduces a novel framework for retinal vessel segmentation, improving accuracy in detecting diseases like diabetic retinopathy. The new method enhances automated analysis of medical images, overcoming limitations of current techniques.
Area of Science:
- Medical Imaging
- Computer Vision
- Ophthalmology
Background:
- Current retinal vessel segmentation algorithms struggle with domain-specific variations and limited training data, hindering accurate detection of ocular diseases like diabetic retinopathy.
- Manual analysis of retinal images is time-consuming, necessitating automated and robust segmentation techniques for clinical applications.
- Existing methods face challenges with unlabeled data, requiring manual intervention and limiting scalability.
Purpose of the Study:
- To develop a novel framework for semi-supervised domain adaptation and contrastive pre-training to improve retinal vessel segmentation.
- To address limitations in current algorithms, including domain shifts and the scarcity of labeled data.
- To enhance the robustness and clinical applicability of automated retinal image analysis.
Main Methods:
- A novel framework combining semi-supervised domain adaptation and contrastive pre-training was developed.
- The model utilizes a pseudo-labeling approach and feature-based knowledge distillation within a temporal convolutional network (TCN).
- The approach focuses on extracting robust, domain-independent features for enhanced cross-domain adaptation.
Main Results:
- The proposed model achieved outstanding performance on the DRIVE and CHASE_DB1 datasets, outperforming state-of-the-art methods.
- Key performance metrics included high accuracy (e.g., 0.9792 on DRIVE), sensitivity, specificity, and AUC.
- Extensive ablation studies and sensitivity analyses validated the model's robustness across different data percentages and parameters.
Conclusions:
- The novel framework significantly enhances retinal vessel segmentation accuracy and robustness, particularly in the presence of domain shifts and limited labeled data.
- The developed semi-supervised domain adaptation and contrastive pre-training approach offers a scalable and practical solution for automated analysis in clinical settings.
- This research advances automated medical image analysis, providing a more efficient tool for diagnosing and monitoring ocular diseases.
