AI meets cardiac medicine: how the University of Siegen wants to make long-term ECGs more reliable
One heartbeat follows the next on the practice monitor: 24 hours of long-term ECG, around 100,000 recorded heartbeats - and only eight minutes of paid time for the doctor to assess them. Is the patient's heart beating in time, is it stumbling, racing, skipping beats? The recording can be used to detect diseases and symptoms, such as atrial fibrillation, arrhythmia or irregularities. This can save lives. However, according to official statistics, the error rate is higher than many patients realize. On average, it is around 25 percent.
This is precisely where the "FACE" project, in which the University of Siegen is involved, comes in. Together with project partners GETEMED and the Berlin Institute of Health (BIH), the "Medical Informatics and Graph-based Systems" research group led by Prof. Dr. Kai Hahn and Dr. Christian Weber is developing and analyzing AI systems to support doctors in the challenging task of evaluating cardiac data quickly and reliably. The total project budget is just under seven million euros, with the University of Siegen receiving around 430,000 euros in funding.
Real patient data provides the basis for the research: The Siegen research group obtains ECG training data free of charge from the renowned MIT (Massachusetts Institute of Technology) in the USA, among others, which has been released for research and has already been expertly evaluated. These data sets serve as training material for various AI systems. The researchers systematically compare their performance: how quickly do the systems deliver results? How high is the computing effort? And above all, how precise are the ECG analyses?
The best-performing systems are then further developed using internal project data sets. The artificial intelligence learns to recognize noise and interference, as well as to classify various cardiac arrhythmias, such as atrial fibrillation or atrial flutter, and the normal sinus rhythm - completely independently, without human assistance.
In the project, the University of Siegen is cooperating with five partners, including manufacturers of ECG devices and pacemakers. The Charité hospital in Berlin is also involved. A feasibility study with the practice partners will show how medical professionals imagine working with AI and what prior knowledge can already be drawn on. What is already clear: Only a data protection-friendly variant is acceptable when using AI. Patient data should be secure at all times. Prof. Hahn and Dr. Weber's team are therefore pursuing an edge solution for use in clinics and doctors' surgeries. This means that the AI reads and analyzes the ECG data directly on site. Sensitive patient data does not leave the clinic or practice and is not sent abroad or to third-party providers - as would be the case with cloud solutions on external servers. This makes edge systems not only much more data protection-friendly, but also independent of network access.
Suitable hardware is required for such edge systems, in this case a laptop into which the ECG data can be fed and on which the AI reads the data. In discussions with the practice partners, it became clear that for the AI to actually be used in everyday practice, these laptops had to be affordable. "If we need hardware costing several thousand euros for the readout, and this laptop is to be used exclusively for reading ECGs, most practices simply can't afford it," reports Prof. Hahn. The aim of the research group is to develop a system that works on hardware that costs around 1,000 euros. The technical architecture therefore plays a central role in the research. The AI should only require as much computing power as a standard laptop can deliver.
The potential benefits are enormous - for medical professionals and patients alike. AI analysis is already more reliable than human evaluation. Abnormalities are detected more precisely, and in a fraction of the time. "Of course, AI systems should only support doctors. A human must always check the AI analysis," explains Jasmin Freudenberg, a doctoral student from the research group. At the moment, AI is not yet able to derive specific suspected diagnoses from abnormalities. However, Annika Steiger, also a doctoral student in the "Medical Informatics and Graph-based Systems" research group, believes that this could also become a reality in the medium term.
Cloud solutions for other target groups are also being developed by the cooperation partners GETEMED and the BIH. The advantage here: More performance. The aim at the end of the project is to use edge and cloud algorithms to design two different solution approaches that make optimum use of the respective advantages of the individual areas.
As the "FACE" project is pre-competitive research, the results may not directly lead to a market-ready product. However, the companies involved want to incorporate the knowledge gained into the further development of their devices. A smart ECG device that does not require an additional laptop is conceivable, for example.
The overall "FACE" project (research into the application of cloud and edge computing for efficient AI-supported ECG analyses) is funded by the former Federal Ministry for Economic Affairs and Climate Protection. The project sponsor is the VDI Technology Center, Düsseldorf. The project is scheduled to run from 2024 to 2026 - with the aim of making cardiac diagnostics more sustainable, precise and safe.
Cooperation partner:
- BIOTRONIK Vertriebs GmbH & Co. KG
- Charité (University Medicine Berlin, Berlin Institute of Health)
- Evangelisches Diakonissenhaus Berlin Teltow Lehnin - Foundation under civil law
- GETEMED Medizin- und Informationstechnik AG
- SEMDATEX GmbH
Photo (from left): M.Sc. Jasmin Freudenberg, M.Sc. Annika Steiger, Dr. Christian Weber and apl. Prof. Dr.-Ing.
