A new autonomous robotic system combines two steady-hand eye robots and three deep learning algorithms to guide needle motion for retinal vein cannulation, validated on pig eyes under static and breathing-mimicking movement.

Developed at Johns Hopkins, the system uses surgical microscope images and optical coherence tomography to autonomously perform retinal vein cannulation with high precision.

Retinal vein cannulation is part of a spectrum of treatments for retinal vein occlusion, alongside anti-VEGF injections and steroids, but RVC requires ultra-precise needle insertion into retinal veins.

Lead author Peiyao Zhang notes the importance of embedding expert surgical principles into deep learning models to enable robot-assisted autonomous procedures that match experienced surgeons’ outcomes.

The hardware-software integration enables precise needle control, with deep learning tracking and planning guiding injections.

Performance metrics show the system completed RVC in 90% of static pig eyes and 83% of moving pig eyes, and reliably detected vein contact and entry.

If validated further, this approach could reduce surgeon workload and increase precision, but must undergo live animal testing and human clinical trials before clinical use.

Future steps include testing the workflow in live animal studies and pursuing translation to real-world surgical settings.

Findings were published in Science Robotics in 2025 (doi: 10.1126/scirobotics.adw2969), with additional editorial context provided by Science X Network.