iOCT-guided simulated subretinal injections: a comparison between manual and robot-assisted techniques in an ex-vivo porcine model

**Background:** Subretinal injection is a key technique for delivering therapies such as gene therapy and stem cells to treat retinal diseases like age-related macular degeneration and inherited RPE dystrophies. The procedure is challenging due to the retina's thinness (~250 µm) and physiological hand tremor (100–200 µm). Intraoperative OCT (iOCT) provides real-time visualization, but manual injection still risks RPE damage, reflux, and inconsistent bleb formation. Robotic assistance may improve precision and outcomes.

**Methods:** Eighty-four ex-vivo porcine eyes were obtained from healthy pigs aged 5–7 months. After excluding 12 eyes (8 cloudy corneas, 3 pre-existing retinal detachments, 1 damaged needle tip), 72 eyes were included. Two specially trained investigators (STI) with no prior vitreoretinal experience performed either manual or robot-assisted subretinal injections (21 per surgeon per method) after standard pars plana vitrectomy. The robot was a custom hybrid parallel-serial device with five degrees of freedom, controlled via a 3D mouse. PFCL (C8F18) was used to simulate drug deposition because it remains localized and is easily visualized on OCT. Injections were performed using a 40-gauge metal bent micro-needle (80 µm diameter) under iOCT guidance (OPMI LUMERA 700 with RESCAN 700). The assistant tracked the cannula in real-time with two perpendicular OCT B-scans. Outcome parameters included successful subretinal blistering (defined as bleb above RPE, no RPE penetration, no bleb rupture upon retraction), reflux, RPE damage, cannula entry angle, achieved tip depth (as percentage of retinal thickness), intraretinal tunnel length, and operation duration. Statistical analysis used two-tailed Chi-square and independent t-tests; p<0.05 was significant.

**Key Results:** Robot-assisted surgery achieved successful blistering in 73.7% of cases versus 61.8% manually (p=0.279, not significant). Reflux occurred significantly less often with the robot: 23.7% vs 58.8% overall (p<0.01), and among successful cases only 14.3% vs 66.7% (p<0.001). RPE damage was observed in 10.5% of robotic and 26.5% of manual cases (p=0.079). Mean cannula entry angle was 44.6° (robotic) vs 39.2° (manual). Achieved tip depth (successful trials without RPE penetration) was 83.1% ± 7.9% (robotic) vs 87.2% ± 5.7% (manual) (p=0.107). Intraretinal tunnel length was 355.5 µm ± 149.4 µm (robotic) vs 427.9 µm ± 124.7 µm (manual) (p=0.144). Operation duration was significantly longer for robotic procedures: 120 s ± 65 s vs 32 s ± 17 s (p<0.001). No learning curve was observed for robotic injections, while manual injections showed an 83% learning percentage.

**Clinical Implications:** This proof-of-concept study demonstrates that robot-assisted subretinal injection with iOCT guidance is feasible and may reduce reflux, a critical factor for cost-effectiveness when using expensive gene or cell therapies. The significant reduction in reflux (from 58.8% to 23.7%) suggests robotic assistance can improve drug delivery efficiency. Although success rates did not reach statistical significance, the trend favors robotics, and the lower RPE damage rate (10.5% vs 26.5%) indicates improved safety. The longer operation time (120 s vs 32 s) is a trade-off that may be acceptable given the potential benefits. Limitations include the use of ex-vivo porcine eyes (which may differ from human eyes in retinal adhesion and mechanical properties), the use of PFCL (a hydrophobic liquid not fully representative of hydrophilic gene vectors), and the inexperience of the surgeons. Further studies with experienced surgeons, human donor eyes, and in-vivo models are needed. The combination of robotic stability and iOCT depth information may also benefit training and enable precise reproduction of injection trajectories for sequential delivery of BSS and therapeutic agents through the same retinotomy.