Abstract
Purpose :
To utilize a combination of a denoising algorithm with an automated OCT segmentation algorithm to quantify intraretinal fluid (IRF) and subretinal fluid (SRF) volumes in patients with retinal vascular diseases.
Methods :
Swept-source OCT was used to acquire volume scans of the macula (Triton Topcon Healthcare, Tokyo, Japan) in patients with diabetic macular edema. Volume scans were processed with Topcon’s denoising algorithm set at 75%. Original and denoised images were then analyzed using an automated OCT segmentation algorithm which provided IRF and SRF volume measurements, which were compared to assess impact of denoising. Visual acuity (VA) was also collected. Fluid volumes were correlated with 6-month change in VA using the Pearson Correlation Coefficient (PCC); the difference in correlations was examined using a Wald test.
Results :
Forty eyes were included in the final analysis. There was a significant difference between paired IRF volumes measured from denoised and original images (P<0.05). Paired SRF volumes were not significantly different (P > 0.05). The PCC between baseline IRF and 6-month change in VA revealed no correlation for the original images 0.07 (P = 0.649) or the denoised images 0.15 (P = 0.343). The difference between PCC values for denoised and original image baseline IRF measurement correlated to change in VA was significant (P<0.05). Baseline SRF was not correlated with the 6-month change in VA on original or denoised images, with PCC of 0.020 (P = 0.896) and 0.019 (P = 0.911), respectively. The difference in PCC between the denoised and original images for baseline SRF and change in VA was not significant (P>0.05). Subgroup analysis considering high and low image quality revealed no significant difference in impact of denoising between these groups (p>0.05).
Conclusions :
Denoising led to a significant difference in the IRF volume measurement. Denoising has potential to improve weak OCT signals, as seen with the significant improvement in correlation between IRF and 6-month change in VA. Denoising does not improve correlations where no signal is present, as seen with the correlation between SRF and 6-month change in VA.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.