DMSU-Net++: A dual multiscale retinal vessel segmentation method based on improved U-Net+.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Accurate segmentation of retinal blood vessels is challenging due to diverse sizes, shapes, and complex capillary structures.
• Existing methods struggle with the fine morphological details and scale variations present in retinal vasculature.

Purpose of the Study:

• To propose an improved retinal vessel segmentation method, DMSU-Net++ (Double Multiscale U-Net++), to address the limitations of current techniques.
• To enhance the accuracy and robustness of retinal blood vessel segmentation, particularly for fine capillaries and varying scales.

Main Methods:

• Developed DMSU-Net++, an enhanced U-Net++ architecture incorporating a novel multiscale feature extraction module (WTSAFM) utilizing wavelet transform.
• Implemented a dual multiscale feature extraction module by cascading MFE modules to capture both frequency and spatial information effectively.
• Evaluated the method on two public datasets: DRIVE and CHASE-DB1.

Main Results:

• DMSU-Net++ achieved an F1-score of 82.75% on DRIVE and 82.81% on CHASE-DB1.
• The method demonstrated high Sensitivity (83.74% on DRIVE, 85% on CHASE-DB1) and AUC (97.86% on DRIVE, 98.36% on CHASE-DB1).
• Experimental results indicate superior segmentation performance compared to other existing methods.

Conclusions:

• DMSU-Net++ effectively segments retinal blood vessels, including fine capillaries, by leveraging multiscale feature extraction.
• The proposed method offers improved accuracy and context understanding for retinal image analysis.
• DMSU-Net++ shows significant potential for clinical applications in diagnosing retinal diseases.