PREDICTIVE OCT BIOMARKERS OF RETINAL CHANGES AND VISUAL OUTCOMES IN SILICONE OIL ENDOTAMPONADE IDENTIFIED BY ARTIFICIAL INTELLIGENCE.

Abstract

To quantify retinal layer changes and visual outcomes in eyes with silicone oil (SO) endotamponade for rhegmatogenous retinal detachment using OCT biomarkers and prediction models. Seventy-six eyes with SO endotamponade underwent macular volume OCT at SO insertion and just before SO removal. An automated segmentation tool quantified retinal nerve fiber layer (RNFL), ganglion cell layer + inner plexiform layer (GCL+IPL), other retinal layers, and fluid (SRF, IRF). Eyes with and without macular edema (ME) during tamponade were compared. A random forest classifier explored predictors of categorical best-corrected visual acuity (BCVA) change (improved, stable, worsened). RNFL and GCL+IPL thinned significantly during tamponade (RNFL 35.4 ± 18.1 µm to 26.9 ± 19.1 µm; GCL+IPL 74.7 ± 17.4 µm to 55.8 ± 27.2 µm), while PR+RPE showed mild thickening (59.6 ± 12.6 µm to 61.4 ± 15.0 µm). Other layers and fluid compartments showed no clinically relevant change. ME occurred in 44/76 eyes (57.9%), but retinal layer changes and BCVA outcomes (improved/stable/worsened) were similar in ME and non-ME groups (χ2 p=0.94). In the prediction model, RNFL, GCL+IPL, and PR+RPE were the strongest contributors, whereas SO duration and ME status had minimal impact. SO endotamponade was associated with selective inner retinal thinning and mild PR+RPE thickening. Postoperative visual recovery depended mainly on inner retinal integrity, supporting OCT-derived layer biomarkers as practical tools for prognostication after SO removal.