EAMR-Net: A multiscale effective spatial and cross-channel attention network for retinal vessel segmentation

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate retinal vessel segmentation is crucial for diagnosing various eye conditions.
• Existing automated methods struggle with multiscale information, thin vessels, and unnecessary data removal.

Purpose of the Study:

• To propose a novel residual U-Net architecture for precise retinal vessel delineation.
• To enhance automated eye disorder diagnosis through improved segmentation accuracy.

Main Methods:

• A novel residual U-Net architecture incorporating multi-scale feature learning and effective attention was developed.
• Drop block regularization was employed to prevent overfitting.
• A multi-scale feature learning module replaced skip connections for enhanced feature extraction.
• An effective attention block was integrated into the decoder for precise spatial and channel information.

Main Results:

• The proposed model demonstrated outstanding performance in retinal vessel delineation.
• High sensitivity values were achieved across multiple datasets: 0.8293 (DRIVE), 0.8151 (STARE), and 0.8084 (CHASE_DB).

Conclusions:

• The novel residual U-Net architecture effectively addresses limitations of traditional methods for retinal vessel segmentation.
• The proposed model offers a promising tool for enhancing the efficiency and accuracy of diagnosing eye diseases.