Photo Credit: Palmihelp

The following is a summary of “Validation of a deep learning model for automatic detection and quantification of five OCT critical retinal features associated with neovascular age-related macular degeneration,” published in the March 2024 issue of Ophthalmology by Ricardi et al.

Researchers conducted a retrospective study to create and verify a deep learning model capable of accurately segmenting five retinal biomarkers linked with neovascular age-related macular degeneration (nAMD).

They gathered 300 optical coherence tomography volumes from eyes afflicted with nAMD. These volumes underwent manual segmentation to identify five critical nAMD features intraretinal fluid, subretinal fluid, subretinal hyperreflective material, drusen/drusenoid pigment epithelium detachment (PED), and neovascular PED. Subsequently, a deep learning architecture employing a U-Net framework was trained to segment these retinal biomarkers autonomously. The evaluation was conducted on a separate dataset, with main outcome measures including receiver operating characteristic curves for detection, presented through the area under the curve (AUC) values both per slice and volume, correlation scores, enface topography overlap (expressed as two-dimensional (2D) correlation score), and Dice coefficients.

The results showed that the model achieved a mean (±SD) AUC of 0.93 (±0.04) per slice and 0.88 (±0.07) per volume for fluid detection. The correlation score (R²) between automatic and manual segmentation yielded a mean (±SD) of 0.89 (±0.05). Additionally, the mean (±SD) 2D correlation score stood at 0.69 (±0.04), while the mean (±SD) Dice score was 0.61 (±0.10).

Investigators concluded that the automated segmentation model accurately analyzed five key nAMD features, paving the way for real-world studies, standardized reporting, and reduced bias in clinical assessments.

Source: bjo.bmj.com/content/early/2024/03/14/bjo-2023-324647