Robots and sensors are already being used to support doctors in state-of-the-art operating theatres. Computer scientist Stefanie Speidel has no doubt that artificial intelligence could be applied to take this one stage further – to create intelligent assistance systems. A professor at the National Center for Tumor  Diseases Dresden and deputy director of the Zuse School of Embedded Composite Artificial Intelligence (SECAI), she explains in our interview how this could improve treatments, where obstacles still need to be overcome and why international and interdisciplinary research is so important.

Professor Speidel, your research focuses on robot-assisted surgery. Hasn’t this already been a reality for some time with devices such as the da Vinci Surgical System?

It’s true that the da Vinci System is widely used. It offers improved ergonomics for the surgeon by allowing instruments to be controlled by joystick on a console. As far as the surgical outcome is concerned, however, the expertise of the surgical team remains the decisive factor. My research is about making robotic systems more intelligent – and turning them into genuine assistance systems.

Why is that important?

Post-surgical complications are the third most common cause of death worldwide. Some of these deaths are considered avoidable. Technologies that systematically document surgical knowledge and experience and make this available during subsequent operations could contribute to improving the quality of treatment while at the same time easing the burden on medical personnel.

How might this be achieved in practice?

The operating theatre of the future will be equipped with numerous interconnected sensors and actuators – such as robotic systems, for example. The data will be analysed in real time. Surgeons will receive useful information just when they need it, either from the surgical robot console, by voice prompt or via smart glasses. This can include anything from visual marking of target and risk structures in the patient’s tissue to recommending specific courses of action and predicting complications, or can even involve the robot assisting with tasks itself: for example, it can precisely guide instruments or a camera, ensure that actions are performed safely, or potentially take over individual clearly defined tasks.

What role does artificial intelligence play in this context?

We need machine learning to understand – on the basis of the sensor data, which includes the robot’s video streams and data – what exactly is happening at a given moment during the operation: which anatomical structures are visible, which instruments are being used, what is happening with the tissue and which risks are there? AI can compare the situation with empirical data from previous operations and determine the specific support that is required at a particular moment.

Is your research profiting from the rapid advances in AI that we are seeing in chatbots?

The boom in AI is a huge help. There are very good models that we can use as a basis. The big challenge is to train them for surgical applications. Very little data is available so far as hospitals record only a few operations. And when operations are recorded, experts have to laboriously annotate the videos, providing all the necessary background information. And in every case, strict data privacy requirements have to be observed.

You are currently exploring one possible solution in a project that involves synthetic training data. Can you explain how this would work?

The basic idea is that potential surgical scenarios could be reproduced under controlled conditions. Medical simulation programmes can model specific aspects such as the patient’s anatomy, tissue deformation or instrument movements; their big advantage is that the underlying information and annotations are directly available. That said, they are, so far, unable to reproduce either the complexity of real-life operations or their visual appearance to a sufficiently realistic degree. We want to combine diffusion models with simulations to generate highly realistic images and video sequences that can then help train the AI assistance systems.

If Germany wants to be at the forefront of AI development, it needs to recruit talents and retain them long-term. Are the Zuse Schools helping to achieve this?

Yes, through SECAI and the DAAD we are able to recruit top young talents from around the world. One unique feature of the Zuse Schools is that they pursue highly interdisciplinary work and combine all kinds of different fields of research. That’s extremely attractive to international scientists.

What are the obstacles when it comes to transferring university innovations into clinical practice?

The highly exacting requirements that need to be met to validate an application for practical use are the sticking point. This is something I experienced during my CoBot research project. Three years ago, we already had an algorithm that worked well for robot-assisted rectal surgery. However, to be able to test it during real-life operations as part of a clinical study, we had to meet a considerable number of the regulatory requirements for a medical product. Companies have entire departments devoted to such processes, but universities don’t have the resources to do so – that can be frustrating at times.

Do you sometimes regret having chosen to work in the heavily regulated medical sector?

No, not at all. I had considered studying medicine but in the end opted for a more technology-related degree. My PhD already gave me the opportunity to combine the two. I really enjoy interdisciplinary work: I firmly believe that different perspectives are needed to find solutions to major challenges. Sometimes this may take a bit more effort, but it’s very satisfying to work on systems that have the potential to improve medical care.

Steffen Ermisch (15 September 2026)