This week, we spoke with Thomas Dratsch, a radiologist in the Department of Diagnostic and Interventional Radiology at the University of Cologne in Germany. Dratsch’s excitement for artificial intelligence stems from a desire for efficiency and optimizing workflows. As a psychologist, he has also developed an interest in the collaboration of radiologists and AI, as well as AI’s influence on radiologists’ performance. Join us as we dive into Dratsch’s mind and perspective on AI in radiology.
What is your background/experience with artificial intelligence and what first attracted you to the topic?
As a radiologist, my interest in artificial intelligence (AI) was sparked by its potential to streamline daily workflows and enhance report generation efficiency. I have conducted research on AI’s role in improving non-interpretive tasks, such as identifying errors in radiological reports and incorrectly labeled X-ray studies. Additionally, as a psychologist, I have studied how radiologists and AI collaborate, particularly focusing on how incorrect AI suggestions can influence radiologists’ performance – a phenomenon known as automation bias.
What are the biggest challenges to AI adoption in clinical practice?
The biggest challenges to AI adoption in clinical practice include the limited scope of current AI solutions, which can only detect a small fraction of findings in a given study. Consequently, radiologists must still evaluate the remainder of the study for undetected findings. Furthermore, radiologists need to assess the AI’s suggestions for accuracy, providing additional cognitive load to their already demanding workflow.
Give us an example (or an educated guess) of what you think AI will be able to do in 3 years. What about in 10 years?
In the near future, AI will ideally assist radiologists with non-interpretive tasks such as measurements, selecting study protocols, and refining reports by identifying errors or converting free text into structured formats. This will allow radiologists to dedicate more time to reading and interpreting studies. For interpretive tasks, AI could act as a safety tool, detecting rare diseases or often overlooked findings. In the distant future, AI’s capabilities will likely expand to detect a broader range of diseases; however, I do not foresee AI independently reporting studies without radiologists’ involvement.
As a radiologist, what is your advice for younger colleagues wanting to dive into the topic of Artificial Intelligence, Machine/Deep Learning, and/or Radiomics? What are your tricks for staying up to date in this fast-evolving field?
The best ideas often arise from the challenges we face in our daily work, with recurring problems serving as great motivators for innovation. Beyond reading research papers, staying updated through social media platforms can be highly beneficial. These platforms allows researchers to share insights, discuss their own work, and respond to criticism, offering valuable perspectives on the latest developments in the field.
