Evaluation of Deep Learning-Based OCTA Denoising in Retinal Vessel Assessment

Area of Science:

• Ophthalmology
• Medical Imaging
• Biomedical Engineering

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss, necessitating accurate monitoring of retinal vasculature.
• Optical coherence tomography angiography (OCTA) provides non-invasive visualization of retinal microvasculature but can be affected by noise.
• Advanced image processing techniques are crucial for improving OCTA image quality and diagnostic accuracy.

Purpose of the Study:

• To evaluate the performance of the N2V2 denoising algorithm for enhancing optical coherence tomography angiography (OCTA) retinal vasculature quantification in patients with diabetes.

Main Methods:

• OCTA scans from diabetic patients were acquired and processed using the N2V2 algorithm.
• Image quality was assessed subjectively by ophthalmologists and objectively using metrics like contrast-to-noise ratio.
• Diagnostic interpretability and quantitative reproducibility of vascular features were evaluated before and after denoising.

Main Results:

• N2V2 significantly improved subjective vessel visualization and objective image quality metrics (contrast-to-noise ratio, peak signal-to-noise ratio, etc.).
• Denoising maintained detectability of microaneurysms and venous beading but led to decreased vessel density measurements.
• Changes in foveal avascular zone metrics and peripheral nonperfusion area were observed post-denoising.

Conclusions:

• The N2V2 algorithm effectively enhances OCTA image quality and visualization, potentially supporting more accurate retinal change assessments in diabetic patients.
• This denoising technique shows promise for advancing the routine clinical application of OCTA by improving feature quantification.