AI instead of HI? —The view from current practice

A frequently mentioned benefit of using AI in radiology is that AI increases the quality of diagnostics, particularly for young and inexperienced radiologists. AI ensures that rookies miss nothing, highlights abnormal features, and provides them with valuable knowledge from literature and research.

Sounds pretty good. But what does this comfort actually mean for human intelligence, HI? If AI can provide (more) reliable results compared to HI, then do we really have to mistreat the brains of students with the pointless acquisition of certain areas of knowledge and train them in identifying patterns? Should their reasoning capacity not then be better used for learning other skills? Reinforcing intercultural and communicative expertise or the correct use of digital solutions to improve treatment, for example. There is an undeniable need for these skills.

But take care. There is a hitch: Regardless of how well AI solutions can read and analyze images, the responsibility for a diagnosis still lies with a human. This means that HI will still have to cover the entire suite of radiological skills and expertise. If you don’t know something, you cannot make a decision and cannot be responsible. Humans can make mistakes and may not be as sharp after a 14-hour shift. But they must be able to recognize their mistakes or inattentiveness when someone points it out to them. And this recognition requires knowledge.

Considering the temptations of “no need to know anything” associated with AI, it is more important then ever to prioritize key radiological skills and expertise in radiological training. In teaching these skills, IT generally and AI specifically will still make an important contribution, however. Young radiologists will (have to) learn how they can use AI for the benefit of the patient and what added value AI brings—and its limitations. They will gather experience in an interplay between HI and AI—potentially raising treatment to a whole new level.

AI systems should be implemented gradually in radiology to ensure that systems are accepted and optimally exploited by radiologists. It is important that it is explained to employees in hospitals how AI systems function so as to eliminate fears and concerns. AI systems for radiology should be developed in close collaboration with radiologists to ensure that the systems meet the needs of users.

It is also important that the development and use of AI in radiology is transparent and ethically justifiable. The introduction of AI in radiology should be accompanied by a comprehensive evaluation to identify its effects on the quality of diagnostics, patient safety, and the job satisfaction of radiologists.

In short

AI can improve radiology but it in no way replaces human intelligence. Radiologists must therefore still have and acquire a high degree of knowledge and expertise about how to exploit AI sensibly. It is also the duty of the manufacturer to provide readily available assistance for this process.