Evaluation of Deep Learning-Based OCTA Denoising in Retinal Vessel Assessment
Translational Vision Science & Technology · 2025
DOI: 10.1167/tvst.14.12.10 ↗Kurz gefasst
Fachartikel in der Zeitschrift Translational Vision Science & Technology, 2025. Verfasst von Zehua Jiang und 5 weiteren Autor:innen. Beteiligte Einrichtung: Medical Retina Department, Singapore Eye Research Institute, Singapore National Eye Centre, Singapore, Singapore. Hauptthemen laut MEDLINE: Deep Learning, Retinal Vessels, Tomography, Optical Coherence, Diabetic Retinopathy. Die Arbeit nennt 31 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Purpose
To evaluate the performance of the N2V2 denoising algorithm on optical coherence tomography angiography (OCTA) retinal vasculature quantification.
Methods
OCTA scans (3 × 3, 6 × 6, and 21 × 21 mm) were acquired from patients with diabetes using Topcon Triton before/after denoising. Perceived image quality was assessed by three ophthalmologists on a 5-point scale (capillary details in the superficial capillary plexus [SCP] and deep capillary plexus [DCP] of 3 × 3 and 6 × 6 mm scans, and vessel continuity and nonperfusion boundaries in 21 × 21 mm scans), and by objective metrics including contrast-to-noise ratio, peak signal-to-noise ratio, edge preservation index, and structural similarity index measurement. Diagnostic interpretability was evaluated through annotation of microaneurysm and vascular beading counts, foveal avascular zone margins, and peripheral nonperfusion borders. Quantitative reproducibility was assessed for vessel density, foveal avascular zone metrics (area, perimeter, circularity), and peripheral nonperfusion area delineations.
Results
We included 145 eyes (mean patient age, 63.97 ± 9.25 years; 40.26% female) (35.17% no diabetic retinopathy, 33.10% nonproliferative diabetic retinopathy, and 31.72% proliferative diabetic retinopathy). Subjective vessel visualization improved in 3 × 3 (ΔSCP, 1.66 ± 0.39; ΔDCP, 1.90 ± 0.43), 6 × 6 (ΔSCP, 1.68 ± 0.36; ΔDCP, 2.11 ± 0.33), and 21 × 21 mm scans (Δvessels, 1.79 ± 0.37; Δnonperfusion, 1.61 ± 0.38) (all P < 0.001). Denoising objectively increased the contrast-to-noise ratio from poor (1.33-1.68) to moderate (2.90-3.97) (P < 0.001), and the peak signal-to-noise ratio (16.74-18.53 dB), structural similarity index measurement (0.60-0.72), and edge preservation index (0.71-0.79) reached moderate levels. Diagnostic interpretability showed most microaneurysms (64.17% SCP, 87.32% DCP) and venous beading (98.76%) remaining detectable. Vessel density decreased in 3 × 3 mm (ΔSCP, -5.81%; ΔDCP, -4.18%) and 6 × 6 mm scans (ΔSCP, -11.89%; ΔDCP, -5.23%; P < 0.001). The foveal avascular zone area increased slightly (Δ+0.01 mm²; P = 0.02), the perimeter expanded (Δ+0.53 mm; P < 0.001), and circularity decreased (Δ-0.16; P < 0.001), with similar peripheral nonperfusion area (Δ+4.84 mm²; P = 0.49).
Conclusions
N2V2 enhances OCTA quality and visualization, which may support more accurate retinal change assessments. Translational Relevance: The N2V2 denoising algorithm can potentially advance the routine clinical use of OCTA by improving the visualization and features quantification.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Autor:innen und Institutionen
- Medical Retina Department, Singapore Eye Research Institute, Singapore National Eye Centre, Singapore, Singapore.
- Duke-NUS Medical School, National University of Singapore, Singapore, Singapore.
- Beijing Visual Science and Translational Eye Research Institute (BERI), Eye Center of Beijing Tsinghua Changgung Hospital, School of Clinical Medicine, Tsinghua Medicine, Tsinghua University, Beijing, China.
- Department of Ophthalmology, Peking Union Medical College Hospital, Peking Union Medical College, Chinese Academy of Medical Sciences, Beijing, China.
- Department of Ophthalmology, University of Lausanne, Jules-Gonin Eye Hospital, Lausanne, Switzerland.
- Topcon, Tokyo, Japan.
Zitieren
APA
Jiang, Z., Zhang, C., Sherif, M., Akiba, M., Teo, K. Y. C., & Cheung, C. M. G. (2025). Evaluation of Deep Learning-Based OCTA Denoising in Retinal Vessel Assessment. Translational Vision Science & Technology. https://doi.org/10.1167/tvst.14.12.10
Vancouver
Jiang Z, Zhang C, Sherif M, Akiba M, Teo KYC, Cheung CMG. Evaluation of Deep Learning-Based OCTA Denoising in Retinal Vessel Assessment. Translational Vision Science & Technology. 2025. doi:10.1167/tvst.14.12.10
BibTeX
@article{Jiang2025,
author = {Jiang, Zehua and Zhang, Chenxi and Sherif, Mohamed and Akiba, Masahiro and Teo, Kelvin Yi Chong and Cheung, Chui Ming Gemmy},
title = {Evaluation of Deep Learning-Based OCTA Denoising in Retinal Vessel Assessment},
journal = {Translational Vision Science & Technology},
year = {2025},
doi = {10.1167/tvst.14.12.10}
}
EUVIMED