INTRODUCTION: This study aimed to identify optical coherence tomography (OCT) biomarkers at baseline and after the loading phase (LP) of antivascular endothelial growth factor (VEGF), predictive of 12 months (12 m) morpho-functional outcomes in diabetic macular edema (DME). METHODS: This multicenter, retrospective study involved treatment-naive DME eyes treated with anti-VEGF agents. The OCT volume scans at baseline, after the LP, and at 12 m were analyzed by an artificial intelligence (AI)-derived platform (Discovery OCT Biomarker Detector; RetinAI AG, Bern, Switzerland). Different retinal layer thicknesses and volumes, intraretinal fluid (IRF), subretinal fluid (SRF), and biomarkers probability detection, including hyperreflective foci (HF) were measured. A random forest model assessed the predictive factors for final morphological and functional outcomes. RESULTS: A total of 77 treatment-naive DME eyes from 64 patients treated with anti-VEGF (88.3% aflibercept, 11.7% ranibizumab; mean n. of injections 9.93 ± 3.18) were enrolled. A significant reduction of all the retinal layers, IRF, SRF, and retinal volumes (p < 0.05) after the LP and at 12 m was found. The random forest model revealed that a higher baseline IRF volume was a moderate predictor and a lower outer nuclear layer (ONL) thickness after LP was a strong predictor for a good morphological response at 12 m. Best-corrected visual acuity (BCVA) prediction remained limited due to weaker associations with OCT biomarkers. CONCLUSIONS: AI-derived software showed promise in detecting OCT biomarkers and improving 1-year outcome prediction in DME management. Baseline IRF volume and ONL thickness after the LP were strong predictors of achieving a structural response at 12 m, with overall good model performance.
Parravano et al. (Wed,) studied this question.