CMFNet: a cross-dimensional modal fusion network for accurate vessel segmentation based on OCTA data

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Optical coherence tomography angiography (OCTA) is a key non-invasive imaging technique for retinal vasculature.
• Accurate quantitative analysis of retinal vessel morphology is crucial for diagnosing fundus diseases.
• Current OCTA segmentation methods struggle to integrate 3D volume data and 2D projection maps effectively, leading to segmentation inaccuracies.

Purpose of the Study:

• To develop an advanced OCTA vessel segmentation method that leverages both 3D volume data and 2D projection maps.
• To enhance the accuracy and reliability of retinal vessel segmentation for improved fundus disease diagnosis.

Main Methods:

• Proposed a cross-dimensional modal fusion network (CMFNet) for OCTA vessel segmentation.
• Utilized separate encoders for 2D projection maps and 3D volume data.
• Introduced an attentional cross-feature projection learning module and a cross-dimensional hierarchical fusion module.
• Incorporated high-level semantic weight information to optimize the fusion process.

Main Results:

• The CMFNet effectively fused features from both 3D OCTA volume data and 2D projection maps.
• Experimental evaluation on the OCTA-500 dataset demonstrated superior performance compared to existing methods.
• Achieved state-of-the-art results in accurate OCTA vessel segmentation.

Conclusions:

• The proposed CMFNet significantly improves OCTA vessel segmentation accuracy by effectively utilizing multi-dimensional data.
• This method offers enhanced diagnostic support for various fundus diseases.
• Cross-dimensional fusion is a promising approach for advancing OCTA image analysis.