Abstract
Purpose :
To validate the role of an artificial-intelligence (AI) based quantification of most relevant OCT biomarkers in assessing disease activity and treatment response in eyes affected by center-involved diabetic macular edema (DME).
Methods :
After adequately training the deep learning algorithm using normal and diabetic OCT images, a specific tool for automatically detect and quantify: intraretinal and subretinal fluids (IRF and SRF), hyperreflective retinal foci (HRF), the integrity of external limiting membrane (ELM) and ellipsoid zone (EZ), and the area of retinal exudates (AHE) was developed, trained and then validated versus two blinded experienced examiners. Three-hundred eyes affected by center-involved DME were consecutively enrolled to validate the automatic tool, and 100 DME treated eyes were used to quantify each parameter over time (at least three follow-up examinations). Fluid volumes (IRF and SRF) and AHE were measured for three concentric circles with diameters of 1, 3 and 6 mm (fovea, paracentral ring and pericentral ring). HRF were measured in the central 3 mm, whereas the integrity of ELM/EZ was quantified in the central 1 mm. ICC was calculated for each parameter between the two human examiners, and versus the automated tool.
Results :
The agreement for any OCT biomarkers detection and quantification, at any time point, between operators was complete (ICC: 1.0), as versus the AI tool (ICC: 0.99). In all DME eyes most intraretinal fluid per square millimeter was present at the fovea, followed by the paracentral ring and pericentral ring (p<0.0001). And this was also the case for subretinal fluid (p<0.0001). In the follow-up eyes, at every time point, this observation was confirmed. HRF significantly decreased after treatment (p<0.001), recovey of ELM/EZ integrity was reached (p<0.005) and the area of hard exudates slowly reduced in all rings (p<0.005).
Conclusions :
Accurate, repeatable location and quantification of major OCT biomarkers of DME, namely: individual macular fluids (IRF and SRF), HRF, integrity of ELM/EZ, and area of hard exudates, are currently mandatory to adequately diagnose and prognosticate treatment response over time. A fully validated AI based tool, as reported, allows the clinicians to routinely identify and quantify the most relevant OCT biomarkers offering an objective way of planning and following DME eyes.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.