Machine learning software piques interest of NHS trusts

  • 11 January 2018
Machine learning software piques interest of NHS trusts

At least 15 trusts in England are said to be interested in new machine learning software designed to support the diagnosis of heart disease, which its developer is planning to offer for free to the NHS.

The machine learning algorithm, developed by Oxford-based start-up Ultromics, analyses echocardiogram images for signs of disease.

The system is said to be capable of spotting warning signs that might be missed by a clinician, so reducing the risk of a patient suffering a heart attack or other complications.

Ultromics hit the headlines over the festive period after it was reported its machine learning software could be rolled out to NHS trusts for free starting this summer.

CEO Ross Upton suggested the technology could save the NHS £300 million a year by reducing the number of people who are incorrectly sent for heart surgery, or are otherwise given the all-clear and later suffer a heart attack that requires treatment.

Upton told Digital Health News: “We realised there was a big problem in diagnosing coronary artery disease (CAD), in that when patients get referred for stress echocardiogram, clinicians miss one in five [CAD] patients.

“It seemed strange to me that clinicians weren’t attempting to quantify the images and were just looking at them by eye. We developed a system that could quantify the disease and is able to extract 80,000 measurements from the images, compared to a clinician, who will only look at about 10.”

The algorithm has been trained with over 120,000 echocardiogram images collected by Paul Leeson, professor of Cardiovascular Medicine at the University of Oxford, who developed the underlying technology alongside Upton.

Initial trials of the system have been undertaken at six cardiology centres across the south of England. Data collection is still ongoing and additional hospitals will take part in the study over the next year.

Results of the study, called Evarest, will be released later in 2018. If deemed a success, Ultromics’ technology will be rolled out to NHS trusts for free from the end of summer.

“The reason we’re giving it back for free to the NHS is because it was developed as part of research done with Oxford University and NHS, which enabled the innovation,” said Upton.

“A lot of companies do research in the NHS, then commercialise it in other countries and forget about it. They effectively just use the NHS for their research. We didn’t want to do that. We are commercialising it for other markets, but we’re first and foremost giving it back to the NHS and recognising that they’re a key part in this innovation.”

Upton said that at least 15 NHS trusts had contacted Ultromics showing interest in the software, although he was unable to specify which.

While he distanced himself from claims that the technology could “save the NHS”, Upton suggested that yearly cost-savings delivered by Ultromics could add up to billions in the long run.

He also explained there was no cost associated with training staff to use the system.

“The training is virtually nothing because it’s nothing that a cardiologist won’t already be able to do in their current clinical workflow,” Upton said.

“The software is designed to sit exactly within their clinical workflow, so they don’t have to change anything about their clinical practice.”

Rise of the machines

Efforts are being made to spur the development of machine learning in the medical space, particularly in the areas of clinical trials, diagnostics and illness prevention.

Proponents argue that a key benefit of the technology is consistency. Unlike human clinicians, who possess varying levels of experience can have their judgement impeded by environmental factors, it is argued machine learning can deliver the level of unmarred accuracy only afforded by computers.

“Someone in a community hospital at 8pm on a Friday who has seen 100-odd scans that day might not be making as good a decision as someone who’s at a leading academic centre,” Upton told Digital Health News.

“It’s about making things consistent, and making clinicians more accurate in their diagnoses.”

He said getting patients directly benefitting from the technology was the next step.

“When patients take part in research trials they often say, ‘it won’t benefit me, but it will benefit the next generation’. That’s not true in this case – it is going to benefit these patients, and we need to ensure that we get our system out and working in the NHS.”

Subscribe to our newsletter

Subscribe To Our Newsletter

Subscribe To Our Newsletter

Sign up

Related News

AI platform for arthritis detection secures £1.2m grant

AI platform for arthritis detection secures £1.2m grant

Henley Business School has been awarded an £1.2m grant to develop an AI system for early detection of rheumatic and musculoskeletal diseases. 
Leeds Teaching Hospitals trials AI prostate cancer diagnosis tool

Leeds Teaching Hospitals trials AI prostate cancer diagnosis tool

Leeds Teaching Hospitals NHS Trust (LTHT) is piloting an AI tool to assess whether it can improve prostate cancer diagnosis.
MHRA issues transparency guidance for machine learning medical devices

MHRA issues transparency guidance for machine learning medical devices

The Medicines and Healthcare products Regulatory Agency (MHRA) has published guidance on transparency for machine learning medical devices.

3 Comments

  • I would imagine it is because any new technology, be it a drug, diagnostic or app, requires evidence that it actually works. That is on real people being seen by real doctors. This nice looking tech has little or no evidence to support any of the claims. Without evidence the device will not be widely adopted, and that evidence should include evidence of cost savings.

  • There’s a lot of best practice here, even leaving aside the high ethical standards (all too rare in business) that acknowledges the IP contribution by the NHS. For technology to be easily adopted, it should for easily into clinical workflow, rather than demanding that the clinician change in order to work around the technology. To be popular, it should also automate or assist a highly repetitive but complex task. The surprise in this article is that only a few Trusts have expressed interest: why aren’t they beating the door down?
    Of course, if the technology turns out to be more reliable than the clinician (and presumably keeps on learning from experience), then there must come a point when the clinician is no longer needed at all.
    Whilst it is easy to see how machines can excel at complex pattern recognition within images, we should perhaps pause and reflect that the a very significant component of the art of medicine itself is recognising patterns in the protean forms of symptom clusters and diagnostic tests that lead to a diagnosis, and from there to a therapeutic plan…
    At what point in the future does the doctor become merely the mouthpiece of the machine?

    • The reason Trusts aren’t beating the door down is that there is little interest in making technology based business change. It’s very rare in this industry to see a mass of mega excited clinicians or organisations hammering you for your technology solution. You have to convince them.

Comments are closed.