Multi-scale Bottleneck Residual Network for Retinal Vessel Segmentation.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Precise retinal vessel segmentation is vital for diagnosing ophthalmic diseases.
• Deep learning excels at segmentation using color fundus images, but research on Scanning Laser Ophthalmoscopy (SLO) images is limited.
• Existing SLO segmentation methods struggle to balance accuracy and model complexity.

Purpose of the Study:

• To propose an efficient and accurate deep learning model for retinal vessel segmentation in SLO images.
• To address the scarcity of research and the limitations of existing methods in SLO image analysis.

Main Methods:

• Developed MBRNet, a lightweight U-Net based architecture for SLO image segmentation.
• Incorporated a Multi-scale Bottleneck Residual (MBR) module to enhance receptive field and retain details cost-effectively.
• Utilized an Attention Gate (AG) module to focus on vascular features and reduce noise interference.

Main Results:

• MBRNet demonstrated superior segmentation performance compared to existing methods on two public SLO datasets.
• The proposed model achieves better accuracy with a significantly lower number of parameters.
• The MBR module effectively expands the receptive field while the AG module improves focus on relevant vascular structures.

Conclusions:

• MBRNet offers an effective solution for retinal vessel segmentation in SLO images, outperforming current approaches.
• The model's lightweight design and improved accuracy make it a promising tool for ophthalmic disease diagnosis.
• This research contributes to advancing deep learning applications in SLO image analysis for ophthalmology.