As new NHSX guidance around artificial intelligence is digested in healthcare, Sectra’s Chris Scarisbrick argues that AI should barely even be noticed by the professionals who use it, and that it needs to be made easy to procure.
AI has become a big focus in healthcare. It even continues to make headlines in the fight against coronavirus, with reports of researchers looking to validate AI models that they hope could help predict which patients are most vulnerable to COVID-19.
In the healthcare diagnostic space particularly, there is enormous potential for AI, where it is now being looked to in order to transform the delivery of crucial diagnoses for a whole range of illnesses and patient cohorts.
AI is here for the long term. And with real world application now beginning to take hold at pace, AI design and the manner in which it is procured and implemented could be big factors in determining whether algorithms are successful or whether they could actually become a hinderance.
Whether we are applying it to specialisms in radiology, pathology, cardiology, or any other profession harnessing diagnostic imaging, AI should improve accuracy and efficiency for end users, and it should speed up the work that they do. If it adds any extra steps, or slows down our clinicians in their mission to improve patient care and the patient experience, then it is a wasted exercise.
So, how do we minimise disruption both for our professionals and for the healthcare enterprise? Well, this can very much start before hospitals even acquire AI applications.
Can trusts be expected to manage multiple vendors?
Though the recent national crisis has shown that technology can be deployed quickly in the NHS when urgently needed, the traditional way the NHS procures IT is to go through a lengthy procurement process, in which organisations evaluate several vendors, shortlist the ones that look most promising, go to a best and final, make a five to 10 year commitment, and sign a contract, all before the IT or digital department needs to think about deploying it. It is a familiar scene in the diagnostic space, as much as any part of healthcare technology, and it can be a complex and intensive experience both for trusts and for suppliers.
The NHSX AI Lab released A Buyer’s Checklist for AI in Health and Care in May 2020, as a short reference to assist the decision-making for procuring AI solutions in the NHS. It is a very comprehensive and well thought through document, addressing the challenges we see today for accelerating safe AI adoption.
Although the NHSX document provides a good and clear checklist, it is still quite a burden on both trusts and smaller/emerging AI developers to run through a lengthy procurement process each time.
Does a commercial function in an NHS trust really want to repeat the same process and payment terms with potentially dozens of different AI imaging providers, dozens of different times?
And can AI vendors, who might be start-up companies or small businesses and who might never have navigated NHS procurement processes, manage to get their solution through the front door?
There are opportunities for established suppliers of traditional IT systems to play a role both in following the new NHSX buyer’s guide, and in relieving these burdens and in making AI acquisition easier, and imaging technology providers could certainly help.
Sectra is committed to doing so, in particular by supporting the hospitals that use our imaging systems by conducting some of the due diligence process on AI vendors where appropriate. We want to be a partner in the buyers’ journey and create a dynamic marketplace from which hospitals or professionals themselves can choose from a range of AI apps, safe in the knowledge that those apps meet necessary regulatory standards and comply with information governance standards.
And we are in continual dialogue with our customers to understand what is really important in clinical practice – which AI applications will make the most impact on the lives of imaging and diagnostic professionals.
The aim of this is to remove a lot of the complexity from the process, and to overcome barriers like the limitations of contracting resource, the scalability of IT, and the ways applications interact.
Automatic and invisible – matching workflow
The manner in which applications interact is particularly important to ensuring they fit within clinical workflows.
By using the enterprise imaging system provider as the point of coordination for imaging AI acquisition, more thought can be put into standardisation and maintaining user control over how images are displayed on screen for professionals. Our ethos is to tailor this to a close and ever evolving understanding of workflow on the ground. We can also ensure an AI app is integrated into the picture archiving and communication system (PACS), where radiologists spend 80% of their working lives, before it is even procured.
But in essence, for the most part, AI should be invisible to the user. Whilst AI will not replace our professionals and for the foreseeable future a human will always need to remain in the loop in delivering diagnoses, most of the time our professionals shouldn’t even notice it is there. If you are using an AI application that measures lung lesions – only scans that meet these criteria should be forwarded on to the AI, and should be sent automatically, without the need for human intervention. The AI can then run its analytics in the background and push results back to the PACS to alert the professional to urgent cases and to help them prioritise and improve the quality and efficiency of their work.
Vendor neutrality
The growth in AI continues at pace. We as technology partners have a responsibility to help hospitals acquire the best applications to meet their needs. This is not about what AI Sectra can develop, for instance. It is about providing a channel in which the fantastic innovation emerging can reach those working hard to support the frontline of healthcare.