The current UK government has made its vision for artificial intelligence use in the NHS very clear. It wants AI, data and innovation to "transform the prevention, early diagnosis and treatment of chronic diseases by 2030", with the UK to be "at the forefront of the use of AI and data in early diagnosis, innovation, prevention and treatment".
Under this vision, AIs could ultimately become the first point of contact for the sick instead of a human doctor, could help healthcare professionals to diagnose medical conditions, and even monitor individuals' health by analysing data from their wearable devices or smart-home sensors.
It's a huge ambition for a set of technologies that are still developing, and whose use is relatively restricted in the health service today. Can AI really make a difference to the future of the NHS?
Certainly there are signs of a growing appetite in the health service for AI technology: around half of NHS trusts are now said to be investing in artificial intelligence in some form.
To date, that adoption has been broadly with a view to achieve one of two aims: to improve clinical practice, using AI to help doctors and other medical personnel with their diagnoses, or to improve administration and support, using AI to streamline day-to-day office management and resource allocation.
Using AI to make the NHS' support and administration functions more effective may be less exciting, but just as necessary. AI can be used for more efficient resource scheduling and dealing with routine admin, helping save money rather than patients' lives. One trial is using AI chat bots as the first line of contact for patients phoning the NHS instead of getting straight through to a human, for example. Thanks to the general underfunding of the NHS and growing demands on its resources, it's these uses of AI that are likely to make the biggest contribution to the NHS in the short term.
On the pure health side, there are also a number of trial projects that use various elements of AI and machine learning. The NHS' most visible uses of AI in this field are those where it's collaborating with Google's DeepMind AI arm: DeepMind has partnered with the likes of the Moorfields Eye Hospital to train an AI to scan images of patients' retinas for signs of eye disease and make treatment recommendations, and learning to assist with radiotherapy planning by working with University College London Hospitals trust.
The use of AI in work with patients will be slower to develop of the two areas, as far more testing will be needed, and various ethical concerns will need working out. However, there are a handful of specific clinical use cases where use of AI is likely to see stronger take-up in the near future, such as reading scans to check for evidence of new disease, or the progression of an existing condition. With enough solid data, AIs should be able to read scans faster than human doctors, and with at least the same level of accuracy.
The only fly in that particular ointment is finding enough high quality patient data.
The NHS would seem to have one clear advantage when it comes to training AI: as the biggest single healthcare organisation in the world, with over 1.7 million staff and one million patients seen every 36 hours, it should have vast stores of data that can be used to train AI. "The UK ranks fourth in the world when it comes to creating the right conditions for a home-grown AI industry to flourish. The decades of big data held by the NHS represents perhaps the UK's single biggest opportunity to advance that position and develop world-beating applications. We see huge opportunities around healthcare outcomes and productivity," Shamus Rae, KPMG partner and head of digital disruption, wrote in briefing note.
With data spanning so many patients and healthcare professionals, you'd be forgiven for thinking that its datasets would be both large and diverse enough to make them the best possible training data for AI algorithms. However, as the healthcare trusts that make up the NHS don't all use the same IT infrastructure, or even all label their data in the same way, datasets are often more limited, and harder to share and integrate, than might be imagined.
"An AI system is only going to be as good as the data it is drawing upon to learn from, and as we understand it, the NHS does not collect data in a systematic way... I think there's this idea that the NHS is this amazing resource of data because we have this nationwide healthcare system, but we're probably quite a long way off having a nationwide set of data that AI could use," Catherine Joynson, assistant director at Nuffield Council for Bioethics, an independent body that examines and reports on ethical issues in biology and medicine, told ZDNet.
What's more, paper-based patient notes are still the norm rather than exception in both primary and secondary care, meaning a mass digitisation effort will be needed before such information can be rendered useful for AIs. AI may itself be able to help fix this problem; one project aims to use machine learning to search through the digitised versions of the piles of paper documents that make up patient's records in order to find the details that doctors need, saving them time which can be spent dealing with patients rather than ploughing through old notes.
While the NHS has the potential to do great things with AI, as is so often the case, it needs to sort out the IT basics first.
In a recent report into AI use in the NHS by think-tank Reform said the NHS needs to move forward with digital plans and increase the interoperability of its current IT systems to make sure that in the future they all stick to open standards, as well as looking at next steps like developing a plan for the integration of new forms of data generated by wearables and sensors at home.
"AI is not the panacea for these back-end implementation challenges and it will not be possible to reap the benefits of this technology at scale if these barriers are not overcome," it warned.
But if the NHS can get the basics mastered, there are clear wins ahead for the use of AI in the NHS. The analysis of information about disease risk factors, progression, and treatment could all benefit from greater use of AI, potentially finding answers where traditional research efforts may struggle.
The decreasing costs, increasing speed, and rising use of genetic sequencing are also likely to drive the greater use of AI in the health service. "Genome sequencing has been moving very rapidly but it's creating such a huge amount of data, we really haven't got the technology to analyse it. We can produce it -- huge sequences -- but we really don't know what to do with it. There's potential there for AI to help with analysis of genetic data analysis, in particular, and help us understand what genetic data means for our health," the Nuffield Council for Bioethics' Joynson said.
Similarly, AIs are likely to have a significant role in the NHS in future analysing population-scale health factors to track and manage epidemics, shape public health decisions, and analyse how lifestyle and environmental factors play into the country's health.
As with any new technology, AI has challenges that must be overcome before the NHS adopts it more widely: for example, there are questions around what level of error the NHS is willing to accept in the diagnosis of conditions, and working out who should be held responsible in the event an AI misdiagnoses a patient.
Clearly, issues around data privacy and patient trust will also be paramount. Medical records are some of the most personal data that exists, and how the NHS handles patient data in the age of AI will be key. "Like any other technology we've seen in the NHS, there's some hype, and there's also some scepticism. With the case of AI, there's a little more of an issue when it comes to showing confidence and trust in the technology. Achieving public trust is of vital importance for the success of AI in the NHS," Dr Panos Constantinides, associate professor of digital innovation and academic director of the AI Innovation Network at the Warwick Business School, told ZDNet.
"When there is a leak somewhere, when the data is not managed appropriately, when there is not consent, people start asking questions, and that hurts the benefits of the technology -- and the technology can do a lot," Constantinides added.
Public attitudes towards the use of AI may have taken a knock when a deal between DeepMind and the Royal Free NHS Trust to share over one million personally identifiable patient records was found to be inappropriate, and the sharing done without proper consent.
Another key challenge ahead for AI is convincing both the budget-holders and clinicians of its benefits, and overcoming their concerns -- worries about the costs from administrators, worries about patient care quality from doctors.
For now, the discussions around deploying AI in a clinical context are chiefly around using artificial intelligence to support doctors in their decision making, rather than to take over the job of diagnosing and treating from them.
However, doctors are well aware of how their job is likely to change with the spread of AI, and seem to be taking a guardedly optimistic stance, hoping it will allow them to spend more time with their patients and less time on routine tasks such as calculating drug dose.
The Royal Society of Physicians, a professional body that accredits doctors, wrote in a recent blog post: "While AI can help physicians, it is unfortunate that some AI proponents do not appear to understand that doctors are much more than diagnostic engines whose skills will be rapidly superseded by AI classifiers.
"Perhaps they have never met the kind of well-trained, experienced general physician who keeps meticulous records and helps patients obtain the best from limited healthcare resources by formulating their needs and preferences, relying as much on bedside as on laboratory or radiographic diagnosis."
Arguably, AI could reshape how both doctors and patients experience healthcare, with artificial intelligence working to advance the health side of the equation, while medical professionals focus on improving care. Until a company comes up with an algorithm for a thoughtful and understanding bedside manner, there is likely to be plenty of room for doctors and AI to co-exist.
RECENT AND RELATED COVERAGE
Surgeons at St Mary's Hospital have been using a HoloLens-based system to pick out blood vessels before surgery.
While more and more of us are wearing fitness trackers, the real benefits of wearables may come from another quarter.
The NHS still relies on paper for many patient records. Getting rid of it could free up time, and money.
The rise of robots is inevitable in healthcare, but for now, keeping it simple is just what the doctor ordered.
Against a background of growing enterprise adoption of virtual reality, mixed reality and similar technologies are beginning to gain a foothold in the NHS.
The Internet of Things must be one of the most high-profile technology trends of the last five years. Could IoT be the backbone of the NHS of the future?
Physician, reboot thyself! The health service is caught between its creaky past and a shiny future. But change is needed, and fast.
Babylon Health gave its technology the same test aspiring doctors take in the UK. The company says it passed with flying colors.
Some medical technologies that use artificial intelligence might benefit patients but result in a drop in health system revenues. This could make widespread adoption of AI in medicine a tough sell.