3D-FVS: construction and application of three-dimensional fundus vascular structure model based on single image...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computational Biology

Background:

• Fundus microvasculature is clinically observed but typically analyzed in 2D.
• Existing methods are limited by equipment and technology, restricting analysis to planar features.

Purpose of the Study:

• To develop a novel method for constructing 3D fundus vascular structure models from single images.
• To generate hemodynamic characteristics for ophthalmic disease diagnosis.
• To enhance the accuracy of retinal vessel segmentation and disease detection.

Main Methods:

• A novel network framework for fundus vascular segmentation.
• Construction of a 3D vascular structure model using segment length, width, and adjacency.
• Generation of a hemodynamic model and selection of relevant features for diagnosis.

Main Results:

• Achieved 98.63% AUC and 97.52% accuracy in fundus vascular segmentation.
• Demonstrated 95% accuracy in diagnosing ophthalmic diseases using extracted hemodynamic features.
• Outperformed existing models in retinal vessel segmentation.

Conclusions:

• Hemodynamic features are crucial for diagnosing retinal diseases.
• The proposed method offers a novel approach to diagnose retinal diseases by analyzing 2D fundus images with 3D feature extraction.
• The method shows superior performance in both vessel segmentation and disease diagnosis.