Researchers at Johns Hopkins University have developed a new robotic system capable of autonomously performing highly delicate eye surgery. The system is specifically designed to treat retinal vein occlusion (RVO), a serious condition in which a vein in the retina becomes blocked, which can lead to (partial) loss of vision.
RVO is currently treated with, among other things, periodic injections of drugs that inhibit abnormal vascular growth or reduce inflammation. A promising but technically extremely complex treatment method is retinal vein cannulation (RVC). This requires a surgeon to insert an extremely thin needle into the blocked vein with great precision in order to administer medication that dissolves clots or regulates vascular growth.
Precision beyond human limits
The veins in the retina are about as thick as a human hair. This makes manual RVC particularly challenging. The required accuracy is less than 100 micrometres, a level of precision that exceeds the natural motor capabilities of humans. Robotic support can offer a solution here by adding stability and repeatability to the surgical process.
The new system, described in a publication in Science Robotics, combines robotics with deep learning and advanced imaging. It uses images from a surgical microscope and cross-sections of eye tissue obtained via optical coherence tomography (OCT).
‘This work builds on our long-standing interest in addressing the extreme precision and stability challenges of retinal microsurgery,’ says Peiyao Zhang, first author of the study. ‘Retinal vein cannulation requires a level of accuracy that exceeds normal human physiological limits. By combining robotic assistance with deep learning, we demonstrate that an autonomous surgical process with high precision and reproducibility is possible.’
Deep learning as a surgical assistant
The system consists of two so-called steady-hand eye robots that hold a microscopic needle and a surgical instrument, respectively. This hardware is controlled by three deep learning algorithms that have been trained to track the position of the needle, predict movements and plan the optimal actions of the robot.
The researchers tested the system on pig eyes, both in stationary situations and under conditions simulating vertical movements like breathing in living patients. The system proved capable of reliably detecting when the needle made contact with a vein and penetrated it.
In the experiments, the robot successfully performed RVC in 90 per cent of stationary eyes and 83 per cent of moving eyes. According to Zhang, this shows that specialist surgical knowledge can be “embedded” in AI models. ‘This could mean that even clinicians without highly specialised training are able to achieve results comparable to those of experienced surgeons with robot assistance.’
Prospects for eye care
Although the results are promising, clinical application is not yet imminent. Further research in living animal models and ultimately in human clinical studies is necessary. Nevertheless, the researchers see their work as an important step towards further automation of complex eye surgery.
‘Our paper shows that a highly delicate retinal surgical procedure can be partially automated in a safe, accurate, and repeatable manner using robotic assistance and deep learning,’ says Zhang. In the long term, this technology could reduce the workload for eye surgeons, reduce variation in outcomes, and increase precision.
If the next steps are successful, the system could contribute to new treatment options for patients with RVO and possibly other retinal disorders. This research thus underscores the growing role of robotics and AI in ophthalmology and the digital transformation of healthcare.
