Choroidal Optical Coherence Tomography Angiography: Noninvasive Choroidal Vessel Analysis via Deep Learning.

Area of Science:

• Ophthalmology
• Medical Imaging
• Biomedical Engineering

Background:

• The choroid, a highly vascular eye layer, is implicated in various diseases.
• Existing tools lack the precision to analyze choroidal sublayer vessel distribution.
• Effective visualization and analysis are needed for clinical applications.

Purpose of the Study:

• To develop a novel choroidal angiography strategy for evaluating sublayer vessels noninvasively.
• To create a robust segmentation model capable of handling low-quality clinical OCT B-scans.
• To enable detailed analysis of choroidal vessel distribution in disease states.

Main Methods:

• A segmentation model extracts choroidal vessels layer-by-layer from OCT B-scans.
• A cross-domain segmentation approach with an ensemble discriminative mean teacher structure is employed.
• The model is trained to handle variations in B-scan quality for reliable reconstruction.

Main Results:

• The proposed method achieved a 77.28% dice score for choroidal vessel segmentation.
• Noninvasive choroidal angiography successfully visualized vessel distributions.
• Central serous chorioretinopathy patients showed significantly lower vascular indexes across all sublayers.

Conclusions:

• The developed method is the first noninvasive tool for clinical analysis of choroidal vessels.
• The code and training data are released to facilitate clinical adoption.
• This technology aids in diagnosing and understanding choroidal diseases through detailed vascular analysis.