Nonperfused Retinal Capillaries-A New Method Developed on OCT and OCTA

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Nonperfused retinal capillaries (NPCs) are indicators of retinal vascular diseases like age-related macular degeneration (AMD) and diabetic retinopathy (DR).
• Accurate quantification of NPCs is crucial for understanding disease progression and developing effective treatments.
• Current methods for NPC assessment may lack precision and efficiency.

Purpose of the Study:

• To develop and validate a novel deep learning-based algorithm for quantifying nonperfused retinal capillaries (NPCs).
• To evaluate the utility of this method in eyes affected by age-related macular degeneration (AMD) and diabetic retinopathy (DR).

Main Methods:

• Utilized averaged, registered optical coherence tomography (OCT)/OCT angiography (OCTA) scans to create high-definition volumes.
• Developed a deep learning algorithm to denoise OCT/OCTA images and segment NPCs by identifying capillaries without corresponding flow signals.
• Investigated the correlation between quantified NPCs and established clinical features in AMD and DR.

Main Results:

• The deep learning algorithm achieved 88.2% accuracy in segmenting NPCs compared to manual grading in DR.
• Significantly increased mean number and total length of NPCs were observed in AMD and DR eyes compared to healthy controls (P < 0.001).
• NPCs significantly correlated with disease severity markers, including geographic atrophy, neovascularization, drusen volume, and EAA in AMD, and microaneurysms and EAA in DR.

Conclusions:

• A deep learning algorithm effectively segments and quantifies nonperfused retinal capillaries using OCT/OCTA.
• This novel method provides a valuable biomarker for assessing disease status and potentially predicting progression in AMD and DR.