Multi-scale retinal vessel segmentation using encoder-decoder network with squeeze-and-excitation connection and...

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate retinal blood vessel segmentation is vital for diagnosing various eye diseases.
• Existing methods may require manual feature engineering or post-processing.

Purpose of the Study:

• To develop a deep learning model for precise retinal vessel segmentation.
• To improve diagnostic capabilities through enhanced image analysis.

Main Methods:

• An encoder-decoder deep learning network incorporating squeeze-and-excitation connections and atrous spatial pyramid pooling.
• Atrous spatial pyramid pooling captures multi-scale features.
• Squeeze-and-excitation connections adaptively recalibrate channel features.

Main Results:

• The proposed network achieves precise retinal vessel segmentation without hand-crafted features or post-processing.
• Evaluated on public datasets (DRIVE, STARE), the model demonstrated superior performance compared to 12 other methods.
• Visual and quantitative metrics confirmed the model's effectiveness.

Conclusions:

• The developed deep learning method offers a robust solution for retinal vessel segmentation.
• The model shows significant potential for application in clinical medical practices for improved disease diagnosis.