ROSE: A Retinal OCT-Angiography Vessel Segmentation Dataset and New Model

Area of Science:

• Ophthalmology and Medical Imaging
• Artificial Intelligence in Healthcare
• Neuroscience and Retinal Imaging Analysis

Background:

• Optical Coherence Tomography Angiography (OCTA) enables high-resolution imaging of retinal vasculature.
• Automated retinal vessel segmentation in OCTA is challenging due to low capillary visibility and complex structures.
• Lack of publicly available OCTA datasets hinders algorithm development.

Purpose of the Study:

• To create a comprehensive dataset for training and validating OCTA vessel segmentation algorithms.
• To introduce a novel deep learning network (OCTA-Net) for improved retinal vessel segmentation.
• To explore the potential of retinal microvasculature analysis in diagnosing neurodegenerative diseases.

Main Methods:

• Construction of the Retinal OCTA SEgmentation (ROSE) dataset with 229 annotated OCTA images.
• Development of OCTA-Net, a split-based coarse-to-fine network for segmenting thick and thin vessels.
• Evaluation of OCTA-Net against state-of-the-art methods on the ROSE dataset.
• Fractal dimension analysis of segmented microvasculature for disease differentiation.

Main Results:

• The ROSE dataset and source code were released for public access.
• OCTA-Net demonstrated superior vessel segmentation performance compared to existing methods on the ROSE dataset.
• Significant differences in fractal dimension were observed between healthy controls and Alzheimer's Disease patients.

Conclusions:

• The developed dataset and OCTA-Net significantly advance automated retinal vessel segmentation in OCTA.
• Retinal microvasculature analysis shows promise as a non-invasive method for detecting neurodegenerative diseases.
• This work facilitates further research in OCTA-based diagnostics for vision and neurological disorders.