Related Experiment Video
Updated: Jun 10, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
A fundus vessel segmentation method based on double skip connections combined with deep supervision.
Qingyou Liu1, Fen Zhou2, Jianxin Shen1
1College of Mechanical and Electrical Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing, China.
Frontiers in Cell and Developmental Biology
|October 21, 2024
Summary
This study introduces DS2TUNet, a novel fundus vessel segmentation model that improves accuracy for ophthalmic disease diagnosis. The model demonstrates strong performance across multiple datasets, aiding early detection and treatment.
Area of Science:
- Ophthalmology
- Medical Imaging
- Computer Vision
Background:
- Accurate fundus vessel segmentation is critical for diagnosing ophthalmic diseases like diabetic retinopathy and glaucoma.
- Current segmentation methods face challenges with fine vessels, complex regions, and generalization across datasets.
- There is a clinical need for advanced algorithms to improve fundus vessel segmentation accuracy and reliability.
Purpose of the Study:
- To develop and evaluate a novel fundus vessel segmentation model, DS2TUNet, addressing limitations of existing methods.
- To enhance the extraction of local and global features for improved vessel segmentation accuracy.
- To validate the model's performance on diverse public and clinical fundus datasets.
Main Methods:
- Proposed DS2TUNet model combines double skip connections, deep supervision, and TransUNet architecture.
- Employed preprocessing techniques including grayscale conversion, normalization, and histogram equalization.
- Utilized ResNetV1, dilated convolutions, and Transformer in the encoder for feature extraction, with double skip connections in the decoder for refinement.
Main Results:
- DS2TUNet achieved high performance on DRIVE, CHASE_DB1, and ROSE-1 datasets, with F1-scores ranging from 0.8195 to 0.8425.
- The model demonstrated strong accuracy (0.9557–0.9741), sensitivity (0.8071–0.8586), and specificity (0.9713–0.9869).
- Excellent performance was also observed on the clinical central serous chorioretinopathy (CSC) dataset, validating its clinical feasibility.
Conclusions:
- The DS2TUNet model effectively segments fundus vessels, overcoming limitations of previous algorithms.
- The proposed method shows significant potential for aiding early diagnosis and treatment of ophthalmic diseases.
- DS2TUNet offers a robust and feasible solution for clinical applications in fundus image analysis.
