Advancements in Robotic-Assisted Intraocular Surgeries: A New Era of Precision

Retinal vein occlusion (RVO) remains one of the leading causes of vision loss worldwide, with limited therapeutic options that directly address the underlying blockage. But a new study is shedding light on a novel approach—one that may transform treatment precision by handing the scalpel, in part, to machines.

Researchers have successfully developed and tested an autonomous robotic system for retinal vein cannulation (RVC), potentially redefining the surgical landscape for this delicate procedure.

RVC itself is not new. It’s a technically demanding technique where a fine needle is inserted into a blocked retinal vein to deliver therapeutic agents directly to the site of occlusion. But the margin for error is razor-thin: retinal veins are less than 150 micrometers in diameter, and a misstep can result in hemorrhage or irreversible damage. As such, the procedure has remained largely in the domain of elite retinal surgeons, often augmented by robotic assistance to stabilize hand motion. The innovation here, however, is a step further—an autonomous system capable of executing the most intricate parts of the procedure with minimal human input.

At the heart of this advancement are two Steady-Hand Eye Robots (SHERs), co-developed with an intelligent computer vision system. One robot is equipped with a 100-micrometer metal needle for cannulation, while the other maneuvers a medical spatula to support auxiliary tasks. Their movements are guided by three convolutional neural networks (CNNs), which were trained to detect key surgical events—namely, when the needle contacts the vessel and when it successfully punctures the vein wall.

To observe and interpret the surgical field, the system integrates a surgical microscope with intraoperative optical coherence tomography (iOCT), offering real-time cross-sectional imaging. These inputs feed directly into the CNNs, enabling the robots to adaptively respond to the tissue environment and adjust movements accordingly. Crucially, while the system handles the most technically challenging aspects of the procedure, the surgeon remains in a supervisory role, performing lower-risk steps and intervening if necessary.

The study evaluated the autonomous RVC workflow on 20 ex vivo porcine eyes, a well-established proxy in ophthalmic research. Impressively, the system achieved a 90% success rate under static conditions. To simulate the complexity of real-world physiology—particularly the subtle, rhythmic movements of the eye caused by respiration—researchers introduced a sinusoidal motion pattern to six additional porcine eyes. Even under these dynamic conditions, the success rate held at a robust 83%, demonstrating not only the system’s precision but its adaptability to the unpredictable realities of live surgery.

This development arrives at a pivotal time, as the field of ophthalmic microsurgery increasingly explores the integration of robotics and AI. While RVC has long been considered a frontier procedure—limited by steep technical demands—the use of deep learning to predict motion and identify tissue interactions may bring new consistency and reproducibility to a field historically defined by human variability.

From a clinical perspective, the implications are significant. Retinal vein occlusion is currently managed with repeated intravitreal injections of anti-VEGF agents, which treat secondary complications like macular edema but do not resolve the occlusion itself. A reliable method for directly accessing the blocked vein opens the door for more definitive therapies, such as localized thrombolytics or novel biologics, potentially reducing treatment burden and improving long-term outcomes.

Looking ahead, the success of this autonomous RVC model in ex vivo settings lays the groundwork for future in vivo studies, and perhaps even early clinical trials. Challenges remain—particularly in integrating the system into diverse surgical workflows and ensuring its safety profile across patient populations—but the foundation is solid.

By marrying the fine-motor capabilities of robotics with the adaptive intelligence of deep learning, this work represents a compelling stride toward safer, more effective interventions for RVO and beyond.