Healthcare has been one of the most promising testing grounds for artificial intelligence, thanks largely to the vast amounts of data, in the forms of medical records and scans, that these smart systems can analyse. But while there are plenty of AI projects underway, there are still barriers to rolling out the benefits further.
Moorfields Eye Hospital in London has been working with Deep Mind and Google Health to develop an algorithm that interprets scans of the back of the eye, which are known as optical coherence tomography scans.
The impact of this AI-led innovation is potentially revolutionary, says Peter Thomas, director of digital innovation at Moorfields Eye Hospital. The algorithm supports automated interpretation of patient scans and gives hospital staff access to excellent diagnostic information.
SEE: Managing AI and ML in the enterprise: Tech leaders increase project development and implementation (TechRepublic Premium)
Yet despite all this promise, the impact of AI isn’t as wide as it could be, at least not yet.
If you deploy AI in a hospital, you’re using the technology in a place where you already have a department full of clinical experts. Yes, they’ll be able to use the interpretation the AI produces, but they’d probably have come up with a similar diagnostic decision themselves.
Thomas, who spoke at the recent virtual HETT Reset event, says that AI will have a bigger impact when you can apply it to a situation where the level of expertise is different, like in the optometry practice on your high street.
However, that’s a big challenge because, at present, the technical infrastructures to support the use of those algorithms in opticians do not exist.
Infrastructure issues aren’t the only barrier to the development of more effective healthcare treatment through AI. Another key challenge is finding ways to bring together data from multiple clinical sources.
Right now, AI is usually applied to single decisions. Thomas gives the example of diabetic retinopathy screening in his own hospital, where every patient with diabetes gets an annual eye scan that determines the level of follow-up care. “We know that AI can deal with that single workflow pretty well,” says Thomas.
Things get more complicated when hospital staff and their AI-based assistants need to go beyond a single source of data. That’s a big issue, as effective healthcare for most patients relies on more than a single data source and usually involves a complex range of information.
If we fast-forward a few more years, says Thomas, and we anticipate a point at which there are multiple autonomous decision-making systems that might be involved in a single patient’s healthcare journey, then there’s going to be a lot of complexity around how staff are going to implement that information in hospitals and how they’re going to monitor that data effectively.
“Each algorithm will need to be monitored for bias and performance as it changes. And there’s the potential for complex interaction patterns when you have multiple algorithms involved in a single patient’s care,” says Thomas, who says the result is clear: the impact of AI in healthcare could be revolutionary, but we’re not there yet.
“We’re still a distance from being at a point where we can start deploying automated clinical management that goes beyond a single decision or a single interpretation. There’s a lot of work to do in terms of getting the right workforce, expertise and structures within the hospital to support that.”
Other experts agree. James Teo, clinical director of AI and data science, and consultant neurologist at Guys and St Thomas NHS Foundation Trust, joined Thomas at the HETT event and says one of the things his team has discovered through its research work is that “big data is really, really big”.
Automated analysis by AI not only feeds the big-data beast but also sends it off in a new direction.
As people become more aware of automation, their expectations are raised. That hope creates more demand for AI systems, which might be implemented before the key use case around improving patient outcomes is actually identified.
“One fear I have is that the process of operating AI and data-driven technologies is that we’ll create an even greater hunger for data, and we’ll end up spending all our time clicking on menus and checkboxes. And that, I think, is the wrong way to travel. I think we need systems that allow us to capture data in a more human-friendly way,” says Teo.
Moorfields’ Thomas agrees, suggesting the main accelerator for AI in healthcare must be clinical usefulness. He says there’s a tendency for healthcare providers to create AI-based point solutions. Startup companies target particular healthcare problems, but those aren’t necessarily the key issues patients face – and, as result, the tech fails to create benefits.
Teo says the result of this badly though-through deployment process is too many point solutions that need to be managed and maintained – and that’s unfeasible for healthcare organisations, especially when you add-in the risk that the startups that create these point solutions might disappear with their products a few years from now.
The answer, suggests Teo, is to create common platforms, or at least common standards, for handling these point solutions. Vendors need to sign up to these standards and the aim for hospital administrators and tech suppliers alike must be to avoid reinventing the wheel.
Indra Joshi, AI director at digital transformation unit NHSX, says her organisation has plans in this direction. It set up the NHS AI Lab in 2019, a £250m programme that aims to accelerate the safe and ethical development and deployment of AI into the health and care system.
One of the Lab’s key programmes of work is about creating projects that take a problem-focused approach to the healthcare challenges that organisations face, rather than simply focusing on the AI products that currently exist.
“We’ve flipped the traditional approach on its head. We ask, ‘what problems are you facing and how can we take some of those problems and develop a solution?’ And if we fail, that’s OK, because AI might not be the solution to every problem,” says Joshi.
The AI Lab recently worked with Kettering General Hospital to develop a process-automation tool to help staff produce complex situational reports that have to be filled out during the coronavirus pandemic. The system automatically reduces complexity, collecting information from a variety of sources, such as frontline capacity records and patient data, and frees up staff to focus on patient care rather than reporting.
This kind of data-enabled automation goes to show how the technology can boost staff productivity and patient healthcare. While AI can have a huge impact on diagnostics and decision-making processes, the biggest impact for now is likely to be around operational processes – and that’s something to celebrate, too.
“People often get excited about the clinical aspects of what AI can do – people always love to talk about how AI can really help in diagnosis. But actually, there’s a quite a lot of great work happening in the back-end processes,” says Joshi.
This is a syndicated post. Read the original post at Source link .