Deep-learning-based motion correction in optical coherence tomography angiography.

Area of Science:

• Biomedical imaging
• Neuroscience
• Machine learning

Background:

• Optical coherence tomography angiography (OCTA) provides high-resolution microvascular imaging.
• Motion artifacts from limited imaging speed degrade OCTA image quality and quantification.
• Accurate visualization of microvasculature is crucial for research and diagnostics.

Purpose of the Study:

• To develop a deep learning framework for correcting motion artifacts in OCTA images.
• To retrieve and restore microvascular architectures compromised by motion.
• To enhance the visualization and quantitative analysis of OCTA data.

Main Methods:

• A two-subnet deep neural network framework was proposed.
• The first subnet identified and removed motion-corrupted B-scan images.
• The second subnet, an inpainting network, reconnected disrupted vascular networks.

Main Results:

• The framework effectively corrected motion artifacts in OCTA images.
• Microvascular networks in mouse cortex in vivo were successfully recovered.
• Both qualitative and quantitative analyses demonstrated significant image improvement.

Conclusions:

• The proposed deep learning method significantly enhances OCTA image quality by mitigating motion artifacts.
• This approach enables more accurate visualization and analysis of microvascular structures.
• The framework holds promise for improving OCTA-based research and clinical applications.