“The algorithm will see you know..”

Will Artificial Intelligence (AI) replace Radiologists?

A few years back, the so-called ‘godfather of neural networks’, Prof Geoffrey Hinton, made a very provocative statement that kick-started a wave of biblical proportions in radiology.
It was late 2016 when he stated:
“It’s quite obvious that we should stop training radiologists right now; deep learning is going to perform much better than humans in less than five years time.”
It is already 2020, we are nowhere near what Hinton expected and radiologists are still here; however, the damage has already been done. Despite the initial enthusiasm, the seeds of fear are planted.

The buzzwords in every imaging congress around the world are AI, ‘deep learning’ and ‘radiomics’ (or imaging biomarkers if you prefer). Everybody is talking about it, but unfortunately very few know exactly what it means.
It is true that the initial wave of machine learning enthusiasm triggered a booming of artificial intelligence research in medical imaging, bringing along huge amounts of funds for RND, but it has also done considerable harm.
Why?
For two main reasons: Firstly, and most importantly because of over-inflated expectations by stakeholders that don’t actually know what medical imaging is really about, and secondly (and this is also based on my own experience) because many medical students, junior doctors as well as consultants have started to believe that AI is indeed going to replace radiologists and nuclear medicine doctors in the near future. In my opinion this is definitely not going to happen and I’ll explain later. But it has made radiologists skeptic and even unwilling or negative to participate in the process, in an effort to defend their status quo.

Nevertheless, at this point it must be stressed that for other reasons besides misleading statements such as Hinton’s, there is an ongoing crisis in medical imaging staffing and residency programs in Europe and especially in Greece that already threatens the entire national health system. Indicatively there is a 60% reduction in radiology and more than 80% in nuclear medicine residency applications in Greece during the last 5 years. This is attributed mainly due to low incomes and general economic crisis leading to brain drain and ‘scientific refuging’.

Despite all the above, is AI really going to replace medical imaging specialties?

In my opinion the answer is No, for the following reasons:

a. Radiologists and nuclear medicine doctors do not just look at pictures! There is a significant pipeline of actions and considerable workflow in their everyday clinical practice, which obviously goes beyond automated image evaluation.
b. Radiologists also perform interventional and therapeutic procedures, less likely to benefit from machine learning.
c. Doctors will always have the final medical responsibility for any diagnosis or intervention.
d. Algorithms don’t usually fail but if they do they can do it dramatically, so a pair of human eyes should always double check the result.
e. Finally, the autopilot did not replace airline pilots. It just made things easier and probably safer.

Yes, AI is definitely going to change Radiology, but It won’t Replace Radiologists.
Most certainly though, radiologists who use AI will replace radiologists who don’t.

Final thought: AI is also going to affect medical education. More specifically how junior radiologists learn! Theoretically a good algorithm will be able to do the job of an experienced consultant by indicating the suspicious regions of an image, ensuring nothing is missed.