AI – A new opportunity?


As a healthcare provider, you may be quite sure that you will never be unemployed. People get sick, they hurt themselves, and all kinds of aches and illnesses will keep doctors and others in the health business occupied infinitely.

However, in radiology, there are prophecies about the advance of AI and how robots, machines, and software will replace us, as soon as they are effective and accurate enough. Anyone who has seen science fiction knows this scenario, the machines take over and then humans are obsolete. Maybe. Radiology is a perfect area for testing out AI and automation, machine learning, and all kinds of useful gadgets and programs – a lot of our work is standardized, follows a known pathway, and can be linked to databases and image recognition. You may have seen AI tested out here on the Collective Minds platform – and if it is not perfect, it still can be quite impressive. Today, radiologists would say: “AI could be a great aid in our work – but we are still necessary”. I agree!

But let us think of another pathway for AI. What about image quality and improvements in scan time? The last year we have seen amazing image quality, performed in just a few minutes. These techniques (for instance Deep Resolve Boost from Siemens, but also available from other vendors) use deep learning to predict information in all voxels, to filter away noise, and to sharpen images – with a stunning result.

How can you improve your imaging with these techniques? There are 2 basic strategies: 1) In the same scan time as before, improve your images greatly, or 2) With the same quality, reduce scan time significantly. Or maybe the 3rd option is the coolest: Make better images with slightly reduced scan times – the best from each option.

Remember that this is not something that changes the radiologist’s job – but the images that are produced are better and faster created.

I saw an example on LinkedIn recently: A fantastic quality shoulder study with 5 sequences, a scan time of under 5 minutes… One of the comments was: “As a tech I am scared…”
Actually, you could scan 10 patients per hour this way – that is 80 patients per 8-hour shift.

Or could you? Such a production would demand changes in the actual physical layout of a radiology department, and changes in staff – maybe you would want to staff each scanner with several techs or others to help them, to make the workflow satisfy the “hungry scanner”.

There are a lot of issues that could be mentioned, and do not interpret this scenario as something I want to happen – but I am certain that we need to think these matters through. Not in the future, but NOW. You cannot build a new radiology department without optimizing for an effective workflow. You should not plan your staffing based on the speed of an old scanner. There will be expectations of cutting the queue waiting for a scan, there will be demands for more capacity if you buy scanners with these AI techniques.

On the other hand – the images are so excellent and good that you may not need to buy a 3T scanner anymore. The radiologists will become happier and maybe more effective. There is always a trade-off.

The AI technique I have described here is not dangerous for anyone – but as always, we need to adjust and use these improvements to the benefit of our patients, ourselves, and our workplace!

Good luck!

Roar Pedersen, Head of MSK, Unilabs Norway, MSK lecturer & Collective Minds Community Manager


Talk to us