More trust in artificial intelligence: What is needed for this in medicine
- Cooperation between experts from medicine and data science is the most important success factor
- Expanding medical education and training to include AI knowledge
- MedTech start-up deepc and Klinikum rechts der Isar of the Technical University of Munich as a practical example
The use of artificial intelligence (AI) in radiology, as in medicine as a whole, is still sometimes viewed with suspicion. Medical training and work with the support of digital technologies are largely in their infancy. Yet the use of AI provides new opportunities for positive change in medical practice with the help of profound data science. And patients also demonstrably benefit from the synergistic interaction of man and machine. The medtech start-up deepc from Munich and the Klinikum rechts der Isar of the Technical University of Munich show what such cooperation can look like.
Radiology has always been a discipline shaped by technological change. The Radiological activity involves much more than just analyzing radiological images such as X-rays, CT, or MRI scans. Finding a probable diagnosis using a wide range of data points is particularly demanding and time-consuming, requiring the highest level of medical experience and constant concentration.
This is an excellent opportunity for the use of AI, which can support a wide range of applications in radiology practice, from image acquisition and processing to diagnosis or prognosis of disease progression. AI algorithms can use millions of data points to detect details that are easy for the human eye to miss or even virtually invisible. One algorithm will identify a small brain hemorrhage more quickly among thousands of cross-sectional images, another will determine the exact volume of a tumor in seconds. AI also needs neither breaks nor sleep and relieves us of repetitive and time-consuming tasks.
So diagnostics can be faster, more accurate, and less error-prone with AI. Nevertheless, according to a study, more than a quarter of medical students say that AI is one of the reasons why they do not choose radiology as their specialty*.
Many experienced radiologists feel the same way: as a study by the European Society of Radiology* shows, more than half of the respondents worry that the use of AI will worsen job prospects.
Priv.-Doz. Dr. Benedikt Wiestler, radiologist and Head of the Computational Imaging research group at Klinikum rechts der Isar, sees these concerns as unfounded: "The use of AI does indeed shift the focus of diagnostic activity, but this can offer a great opportunity for more concentration on the essentials. For example, only doctors can address questions that may be important for the clear classification of a disease. AI does not explain the diagnosis to patients, it gives them and their relatives neither security nor support. In the future, we specialists will continue to be indispensable, and even more in demand than ever, when it comes to cooperation between different disciplines and the exchange of medical information on complex clinical pictures".
The interplay of medical experience and competence with AI thus enables better concentration on complex cases, thus more patient safety and more time for communication - well-known needs that benefit doctors and patients, and for which there is usually too little time in everyday hospital life.
Physician and data scientist Dr. Franz Pfister, CEO of the MedTech start-up deepc, which has developed one of the leading AI platforms in the field of radiological diagnostics, explains: "With the entry of AI into radiology, the challenge arises to impart necessary data science knowledge at an early stage. General medical training should therefore already include the basics of AI: Physicians should be able to decide how AI algorithms are interpreted and incorporated into clinical decision-making. Understanding terms like machine learning, supervised and unsupervised learning, deep learning, and other concepts are essential, as well as how training data shapes an AI algorithm to provide real clinical value."
Close cooperation between data scientists and radiologists in the testing and evaluation of AI applications is therefore particularly promising.
deepc and Klinikum rechts der Isar started in 2020 with a joint project "NeuroPIPE", funded by the "Bayern Innovativ" initiative, to test automated integration and orchestration of AI algorithms for everyday clinical use.
The integration of artificial intelligence into radiological practice is resource-intensive. “Numerous AI applications on one side and a wide variety of hospital IT systems on the other side, make integration complex and expensive, and the medical users are usually not familiar with AI. Not to mention the regulatory requirements that have to be met, for example concerning the General Data Protection Regulation (GDPR) and the use of algorithms as medical devices," says Dr. Franz Pfister from deepc, summarizing the challenges.
In the joint project with Klinikum rechts der Isar, deepc laid the foundation for the product development of the deepcOS operating system, which has been approved for the European market as a CE-marked medical device.
The system includes an AI platform with a myriad of AI solutions in the field of radiological diagnostics from leading AI companies worldwide. These applications, which can be used for more than 25 clinical indication fields, are all tested, regulation compliant and can be integrated into all common hospital information systems after a one-time, straightforward installation. The hospital thus has the certainty of having access to many AI applications with one platform, and thus also significantly lower costs.
"Without the close and productive cooperation with radiologists, we would certainly not have been so fast and good at developing deepcOS. We were thus able to find solutions that really work in clinical practice later on and, above all, create trust towards the use of artificial intelligence," summarizes Dr. Franz Pfister.
In the meantime, deepc has launched a research project with the University of Landshut and is promoting further scientific validation in the use of AI for radiological diagnostics.
