One of the areas in which artificial intelligence offers the greatest potential is medical diagnostics. Even today, AI algorithms can already detect abnormalities almost as well as human experts can. Professor Daniel Rückert from the Technical University of Munich (TUM) and the Konrad Zuse School of Excellence in Reliable AI (relAI) explains how this works and how such systems provide diagnoses that are in fact reliable. relAI is supported by the DAAD with funding from the Federal Ministry of Education and Research (BMBF).

Professor Rückert, one often reads that AI systems nowadays can analyse X-ray and MRI images at least as well as—and in some cases even better than—specialist doctors. Is that true? And how far have we already progressed in this sense?

It is true that AI systems are becoming better and better at analysing and interpreting radiological images. This is chiefly because the AI is trained on millions of images, allowing the system to learn how to identify even those abnormalities which are difficult to detect. An experienced radiologist can do the same, yet we unfortunately have a shortage of such specialists—not only in Germany, but even around the world.

You have developed new algorithms capable of reconstructing, analysing and interpreting biomedical images. How exactly did you go about improving these systems?

One challenge involved in obtaining MRI images concerns the length of the process. The patient has to lie in the MRI scanner for a long time, ideally without moving. For many patients, this is unpleasant to say the least, and virtually impossible in the case of young children. That’s why we have developed AI systems that can use only a fraction of the MRI sensor information to reconstruct high-quality MRI images. This significantly accelerates the imaging process.

Analysing imaging data is just one of the ways in which AI can be deployed in medicine. What other examples can you give us?

In principle, AI systems can be deployed in all areas of medicine that require data to be interpreted, such as pathology, genomics or laboratory medicine. However, AI can also be used in robot-assisted surgery or for the optimal planning of radiation therapy for cancer patients. In addition, artificial intelligence can relieve the burden on doctors and nursing staff, for example by automatically generating doctor’s reports, discharge letters and care records. Another useful application for AI is the prediction of bed occupancy rates in hospitals, which greatly facilitates planning and makes life easier for nursing staff. 

The increasing personalisation of medical treatment is regarded as one area with huge potential. What options does artificial intelligence offer in this respect?

There is increasing realisation in the medical world that many diseases that were previously viewed as belonging to one specific type actually encompass a very broad spectrum of different variants. They can be divided up into many sub-types, each with different underlying disease mechanisms. This is precisely how dementia – among other things – is viewed nowadays. The challenge now is to correctly identify these sub-types, which is where AI can prove an enormous help thanks to its ability to comb through enormous volumes of data in search of specific criteria.

You are also working on AI-based methods of preventing illnesses and detecting them at an early stage. How exactly does this work? How can clinical information be extracted from medical images?  

Early detection is very important because many diseases are easier to treat and cure in their early phases. However, to achieve this it is essential to identify abnormalities early on that may be a sign of specific illnesses. Here too, AI plays a major role. At the same time, it is very important to ensure that AI-based early warning systems do not give any false alarms, as this can cause psychological stress and lead to unnecessary examinations or treatments. 

Isn’t it risky to have AI systems interpreting medical data? 

The aim should not be for artificial intelligence to do the diagnosing instead of doctors, but for it to support them in the process. Such systems should nonetheless function as reliably as possible, of course. This is one of our research goals at the Zuse School relAI – to develop trustworthy AI systems. With such AI systems, the central focus is on explainability and reliability. 

You teach at the Zuse School relAI, which has defined the development of reliable and responsible AI systems as one of its focus areas. What precisely needs to be considered in this context?

The development of reliable and responsible AI systems is a very exciting area of research. That’s also clear from the huge interest shown by young scientists in the topic. Above all, it is exciting to watch how the development of transparent and explainable AI systems also improves our understanding of basic AI principles and helps us develop new and better systems.

You were recently awarded the 2025 Gottfried Wilhelm Leibniz Prize. What significance does this award have for you personally, as well as for your future research projects?

Naturally I was delighted to receive this award. The Gottfried Wilhelm Leibniz Prize is the highest German research prize, and an honour that means a great deal to me. Especially because this prize is of course also an acknowledgement of the many young researchers – doctoral candidates and postdocs – with whom I have been able to collaborate. The prize naturally also gives us more flexibility when it comes to future research in our group, providing us with the opportunity to explore new and promising directions in the area of AI and medicine.

Interview: Klaus Lüber (6 February 2025)