Radiology residents have strong confidence in the future of diagnostic radiology

What is the future of diagnostic radiology? Will AI keep medical students away from the field? Will they prefer a different specialty because they expect AI and machine learning to replace radiologists completely in the future? A recent study amongst Canadian medical students performed by Gong et al., discussed on concluded that many respondents considered choosing another specialty than radiology because of the rise of AI.1,2 How do radiology residents see the future of radiology? Do they think AI is an exciting opportunity, ready to supercharge the radiologist? Or are they as skeptical as the Canadian students?

To answer this question, we asked attendants of an AI-focused training day for Dutch radiology residents what their thoughts on AI are. Dutch radiologists-to-be show a more positive outlook on the future than Canadian students. When asked whether their “profession will flourish like never before”, all 36 respondents agree. What do the residents think about challenges and expected impact of AI in the field of radiology?3 What are their concerns and what do they hope for? 

Challenges are in lack of knowledge on AI and workflow integration

Although radiology residents expect a bright future, they do not consider clinical implementation of AI in radiology practice a problem with a straightforward solution. Figure 1 gives an overview of expressed concerns. “Lack of knowledge about AI” is number one on the list. Interestingly Gong et al. also advise to improve education for medical students on the potential impact of AI.1 The Dutch residents rank “Unclear workflow integration” the second biggest challenge for AI implementation. Additional comments during the on-site discussion highlighted the complications that are expected when radiologists will have to deal with even more software programs than they have on their desktops now. Efficient communication between the different IT solutions is also expected to be a challenge.

Future of diagnostic radiology chart highlighting the biggest roadblocks to implementing AIFigure 1: Challenges perceived concerning the implementation of AI in radiology practice. Nr. of respondents: 26.4

Other shared worries are around who carries ultimate responsibility for the output of an AI algorithm. Will the radiologist always be ultimately responsible for the diagnosis? Or will software development companies carry the responsibility in the future? This question was answered by 37 attendees, out of which 21 expect radiologists will always be held accountable for decisions made by the algorithm. Although during the discussion that followed, it was questioned how radiologists can trust an algorithm if the company developing the software does not want to be held responsible for the output of the algorithm.3

Additionally, concerns were expressed around collaborating with parties like Amazon, Google or Facebook for storage purposes. Would this mean violating the Hippocratic oath by giving these companies unlimited access to highly personal radiology reports? Of 37 respondents 13 attendees answered this question positively. Remarks from the audience added color to the feeling on the subject: “If Amazon’s business would be and remain purely about storing data, meaning this is what they aim to do best in the most secure way, that would be fine. However, if Amazon starts providing other services, related to healthcare, or insurance, it might become a problem. What if people become uninsurable because Amazon knows too much about their health?”3


Biggest impact expected in speed of diagnosis

The exact impact AI will have shaping the future of diagnostic radiology is still a topic of intense debate. The Dutch radiology residents have highest expectations for speeding up diagnostic workup (>60% of respondents); accuracy improvement comes in a convincing second (>30%). Remarkably, none of the respondents expressed any hope for lower medical costs. Combined with the worry about healthcare costs concerning AI implementation (see figure 1) this reveals little trust in the business case for AI in radiology. See figure 2 for the poll results. Interestingly, the Canadian medical students are also hopeful for more efficient radiology processes. Approximately 75% of respondents agreed with the statement that AI will augment the radiologists’ capabilities and make radiologists more efficient.

Future of diagnostic radiology chart describing how AI will have the biggest impact in radiology in five yearsFigure 2: Expected impact concerning the utilization of AI in radiology practice in 5 years. Nr. of respondents: 26.4

So what have we learned about the future of diagnostic radiology?

AI implementation in the field of radiology is not a done deal. Radiology residents are perfectly aware of this fact and experience a need to expand their knowledge about AI. At the same time hopes are expressed around acceleration and accuracy improvements of the diagnostic process. If we want to make the AI transformation a success, we should focus on enhanced education on AI to equip the upcoming radiologists for the future of diagnostic radiology.

A special thanks to the organizers of the AI-day for radiologists on the 2nd of February 2019; Dr. Ayoub Charehbili, Ben Zwezerijnen, Maarten van de Weijer, and Dr. Merel Huisman. Additionally, I would like to thank Prof. Dr. Tim Leiner, for making his poll results available to include in this article.


  1. Gong, D. X. X., Nugent, P. & Chang, J. Influence of Artificial Intelligence on Canadian Medical Students' Preference for Radiology Specialty : Acad. Radiol. 1–12 doi:10.1016/j.acra.2018.10.007
  2. Brown, D. AI in Healthcare. AI causing anxiety for some medical students (2018). Available at:
  3. Leiner, T. Interactieve discussie en meet-the-expert, AI day for AIOS, UMC Utrecht, 2nd of February 2019. (2019).
  4. Leonaite, J., Six, O.R., van Loon, A. Quantib on-site poll, AI day for AIOS, UMC Utrecht, 2nd of February 2019. (2019).