Prof. Dr. Ramona Woitek heads the Medical Image Analysis & Artificial Intelligence (MIAAI) research group at Danube Private University, Austria. The research center focuses on medical imaging analysis to develop quantitative biomarkers, and investigates to predict the presence of cancer and its progression or response to therapy with AI. In an interview with Mélisande Rouger, Woitek discussed her research and why medical students and radiologists must learn how to use AI correctly.
Do you believe that AI is here to stay in radiology?
Yes, and this will be the great challenge of the future. More and more research papers, lectures, and events deal with AI in medical imaging, as well as companies that want to sell their image analysis algorithms. In the past, radiologists were apprehensive about using AI- at least in Austria. They were worried that AI would take away their jobs, but this attitude seems to have changed. Partly because there is a staff shortage in radiology, which is particularly pronounced in the UK, but is also noticeable in the rest of Europe. Radiologists are now beginning to embrace AI as a possible tool to counter this. AI can help us focus on the diagnosis and the patient, and is primarily helpful for repetitive and error-prone examinations.
More and more radiologists are, therefore, seeing the benefits of AI for their work. What about medical students?
We also teach medical students how to use AI and show them what it can already be used for today and what possibilities may open up in the future, especially because ChatGPT and other large language models are freely accessible. Medical training will change dramatically. We need to make an early effort to train students to use AI, and ensure that we can assess our work without the support of such tools, especially when it comes to writing scientific studies. Students still need to be able to work scientifically and write texts without large language models.
How do you train students in AI at your center?
They learn about text generation and data analysis in medicine using AI, and the advantages and disadvantages using practical questions. We also provide insights into our research. We want them to experience the various possibilities that AI offers in medical imaging, and, of course, we want to arouse interest in taking a closer look at it.
What should they learn?
If they later decide to purchase an algorithm, they should be able to compare different solutions and know which produces the best results for the work they need.
This requires some specialist and statistical knowledge. You have to deal with the quality criteria. Choosing the right AI tool is similar to selecting the proper medication: just as you need to know which medication helps with which illness and what the study situation looks like, you need to know which AI to use for which applications.
What skills will radiologists need to use AI correctly?
I am firmly convinced those who use AI will prevail over those that don’t. We must learn how to use AI carefully and update our knowledge to use the tools gradually coming onto the market safely. Radiologists also need to be able to assess the algorithm’s answers for efficiency and accuracy.
How do you train already working radiologists to use AI?
So far, only to a minimal extent. This is probably because the new technology is not widespread in Austria. However, studies show that radiologists want to use AI to be able to make faster diagnosis.
How will AI continue to develop in radiology?
The technology will gradually find its way into the field. The evaluation of mammograms may be the first area in which AI can be used. Algorithms are already showing excellent results when analyzing MRI scans of the prostate. In the case of very complex diseases, radiologists will still be needed to look at the images. A particular challenge will be the training of radiologists, because they will also need to see many images of healthy people. However, how will radiologists be trained if they no longer do this because the technology rejects inconspicuous scans?
Can you give a specific example from your research?
We are investigating the characterization of different tumors using AI. We can use ultrasound, mammography, and magnetic resonance imaging to visualize tumors, but the information that radiologists can obtain from these examinations when diagnosing is limited. The solution is to extract even more information using other techniques. We use radiomics and machine learning for this purpose. PET examinations with novel tracers and new MRI techniques can also tell us much about the microenvironment in which a tumor is located. We are working on combining the radiomics features from standard imaging and molecular imaging. In doing so, we hope to predict better clinically relevant results, such as the response to new drugs.