Related Experiment Video
Updated: Oct 23, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
SERR-U-Net: Squeeze-and-Excitation Residual and Recurrent Block-Based U-Net for Automatic Vessel Segmentation in
Jinke Wang1,2, Xiang Li1, Peiqing Lv2
1Rongcheng College, Harbin University of Science and Technology, Rongcheng 264300, China.
Computational and Mathematical Methods in Medicine
|August 23, 2021
Summary
A novel SERR-U-Net framework enhances retinal vessel segmentation accuracy using attention mechanisms and residual learning. This deep learning approach shows promising results for clinical assistance in analyzing retinal images.
Area of Science:
- Ophthalmology
- Medical Imaging
- Computer Vision
Background:
- Accurate retinal vessel segmentation is crucial for diagnosing various eye diseases.
- Existing methods face challenges with small vessels and complex vascular structures.
Purpose of the Study:
- To develop an advanced deep learning framework for precise automatic retinal vessel segmentation.
- To improve the accuracy and robustness of retinal image analysis.
Main Methods:
- Proposed a SERR-U-Net framework integrating Squeeze-and-Excitation (SE), residual modules, and recurrent blocks.
- Modified U-Net architecture with recurrent blocks for increased depth and residual modules to prevent vanishing gradients.
- Employed SE structures for attention mechanisms and Enhanced Super-Resolution Generative Adversarial Networks (ESRGANs) for noise reduction.
Main Results:
- Achieved high accuracy on the DRIVE dataset (0.9552) and STARE dataset (0.9796).
- Demonstrated strong performance with AUC values of 0.9784 (DRIVE) and 0.9859 (STARE).
- The method proved effective in handling challenging cases like small blood vessels and intersections.
Conclusions:
- Developed an improved U-Net model combining SE, ResNet, and recurrent technologies for automatic retinal vessel segmentation.
- The proposed model offers superior accuracy compared to existing learning-based methods.
- Validated robustness in segmenting challenging vascular features, indicating significant clinical potential.
