We seem to be in the middle of yet another AI hype cycle, and radiology is usually one of the first areas it’s suggested that AI will be transformative. This technology is incredibly exciting, but do we need a reality check before we commit to real-world AI implementations?

In recent years we have seen huge excitement about the potential for AI to replace radiologists, particularly in the context of the radiology workforce crisis. The Royal College of Radiologists (RCR) have been calling for greater investment to grow the radiologist workforce for years, but these calls have mostly fallen on deaf ears. The radiologist workforce grew by just 3% in 2022, whilst the RCR estimates that there is currently a 29% shortfall of clinical radiologists in the UK.1

Attempts to increase the consultant radiologist workforce are underway, but improvements are simply not keeping up with demand. Last year there were 44 million imaging tests reported in England alone – an increase of 26% on the previous year. This huge pressure on radiology services not only means that patients are waiting longer for their scan results, but also that there is an increased risk of reporting errors. Only 24% of clinical directors currently think that their radiology department has enough clinical radiologists to provide safe and effective patient care.2

The potential of AI is even more attractive because healthcare systems simply cannot function without radiology. There are over 3.6bn radiology exams globally each year, including 43 million in the UK alone. Every medical department relies on radiology to provide diagnoses, and without a diagnosis, treatment is often impossible.

However, the reality is that at the moment, AI tools aren’t really able to do more than support decisions made by radiologists. While there are a number of AI models that are now capable of generating detailed narrative reports, recent research concluded that they are still not performing well enough. So, we’re still in the position of not having a clear use case for AI in radiology and rapid integration is likely to only lead to more delays and frustrated staff.

Another huge issue for the NHS to overcome before it can even consider large-scale AI integrations is the state of its digital network infrastructure. When some hospitals in the UK are still struggling with basic infrastructure like reliable wireless technology, how can we justify the investment that AI integrations would require?

So, let’s hope that there will be a future where AI can work alongside radiologists to enhance efficiency and improve workflow. But, we are nowhere near that point yet. So instead of committing millions of pounds of funding towards AI that we won’t be able to use, let’s put that funding towards improving the NHS’s overall digital infrastructure so that we’re ready and able to embrace innovations when they come.

It feels like we’re on another upward curve in AI at the moment where the enthusiasm and excitement at the prospect of an ‘easy fix’ for the NHS are at an all-time high. Recent discussions about regulation of AI have done little to temper this in the popular imagination but are hopefully giving leaders in AI development pause for thought.

We need to remember that innovations never follow a constant upward trend, and we’re very much in the ‘emerging’ stage for AI in radiology. There will almost certainly be a future for AI in healthcare, but it might not be quite the one we expect.

1 RCR Clinical Radiology Workforce Census 2022
2 NHS England Diagnostic Imaging Dataset Annual Statistical Release 2021/22