Every solution to a problem seems to create new complications. Improved access to healthcare, rising living standards and technological advances have greatly improved our lifespans and contributed to higher populations than ever before.

But rising demand for access to healthcare, coupled with more people living into old age, also places an increasing strain on healthcare providers.

As society becomes ever more mobile, it is important to make sure that patients’ records can follow them to make GPs and hospital doctors aware of their full medical history. But moving patient data from paper to computer is yielding far greater benefits than just improved administration.

In healthcare, we have seen an astounding level of hype surrounding the use of AI – but there is real promise for helping people

Professor David Clifton

Researchers at the University of Oxford’s Radcliffe Department of Medicine have been using machine learning – a form of artificial intelligence (AI) – to examine echocardiograms of patients visiting hospital suffering with chest pain. The new system can detect 80,000 subtle changes that would be otherwise invisible to the naked eye, improving diagnosis accuracy to 90% and potentially saving the NHS millions of pounds in avoidable operations and treatment. This is just one of many new applications of AI to healthcare.

‘Digital technology is now part of everyday life,’ says David Clifton, Associate Professor in Oxford’s Department of Engineering Science.

‘In healthcare, we have seen an astounding level of hype surrounding the use of AI – but there is real promise for helping people. For example, one thing AI can do better than humans is to assimilate enormous amounts of data, continuously, and use this to spot subtle events in patient data that are otherwise easily overlooked.’

Professor Clifton’s team in Oxford is currently working in partnership with several research centres in the UK and in China (where he has an Oxford University laboratory funded by the Chinese government) to create enormous databases of medical data that can be used to develop new generations of complex AI algorithms for healthcare.

‘AI algorithms are certainly “data hungry”, but the Oxford approach is grounded in ensuring that everything we do is driven by medical doctors – it is that clinical knowledge, baked into the algorithms, that separates so-called clinical AI from regular AI,’ says Professor Clifton. This approach is being demonstrated at scale by Sensyne Health, a company based on Oxford University technology formed by Lord (Paul) Drayson, a former science minister in the UK government. The outputs from the labs of Professor Clifton and Professor Lionel Tarassenko are being delivered into the NHS via Sensyne Health.

The current applications of AI aren’t, however, limited to improving diagnosis. Many medical techniques require years of practice to perfect, but some researchers are developing technologies that could enable computers to help to improve the skills of less experienced hospital practitioners.

Alison Noble is the Technikos Professor of Biomedical Engineering in Oxford’s Department of Engineering Science. Her main research interest is in biomedical image analysis, with a particular focus on raising the profile of ultrasound imaging, and she has been developing technology to assist ultrasound scanner operators.

Computers don't see data in the same way humans do

‘Ultrasound machines are complex devices to master,’ says Professor Noble. ‘They involve constant interpretation of the data on screen, which directs the actions of the technician performing the scan.

Computers don't see data in the same way humans do
Ultrasound image of developing baby

‘Computers don’t see data in the same way humans do. While we filter out what we see as noise or static, looking for anything we recognise as a head or a foot, computers can analyse all of the data at once to extract vital clues about what the scanner is actually passing over.’

One of the programmes Professor Noble has been developing is able to recognise the key features that doctors look for in the normal development of babies during routine ultrasound scans of pregnant women. Once the computer has recognised a feature such as the head or a heartbeat, it flags it to the technician who can then move on to look for the next feature.

‘This active assistance from the computer is particularly useful for less experienced practitioners, ameliorating the effects of lower levels of training in remote areas where women may not have easy access to hospitals,’ adds Professor Noble. ‘This technology can be used with a small, portable computer and a handheld scanner, effectively providing patients in remote, rural parts of the world with access to much more accurate healthcare diagnostics than before. 

‘It also reduces the need for repeated scans of the same area, making the process safer for the baby.’

The ability of AI to look through the noise in medical scans is also yielding interesting results in preventative healthcare.

Charalambos Antoniades is Professor of Cardiovascular Medicine at Oxford and leads the Oxford Translational Cardiovascular Research Group. His team has developed a new technology that analyses coronary computed tomography (CT) angiograms and can flag patients who are at risk of deadly heart attacks years before they occur.

‘The standard software currently used with CT scanners is designed to filter out certain types of tissues, such as fat, to make it easier to see organs like the heart,’ he says. ‘However, huge amounts of data are obtained with each CT scan, which are currently not used because we don’t know what they mean. This is what our research brings to the surface and analyses.’

This new technology may prove transformative for primary and secondary prevention

Charalambos Antoniades

Heart attacks are usually caused by inflamed plaques in the coronary artery causing an abrupt blockage of blood getting to the heart. Professor Antoniades’ team has developed a technology, called the fat attenuation index (FAI), which detects the inflamed plaques prone to causing heart attacks by analysing CT images of the fat surrounding the arteries – something that is filtered out by any standard CT image analysis software.

‘This new technology may prove transformative for primary and secondary prevention,’ he adds. ‘For the first time we have a set of biomarkers, derived from a routine test that is already used in everyday clinical practice, that measures what we call the “residual cardiovascular risk”, currently missed by all risk scores and non-invasive tests.

‘Knowing who is at increased risk for a heart attack could allow us to intervene early enough to prevent it. I expect these biomarkers to become an essential part of standard CT coronary angiography reporting in the coming years.’

In common with the other research teams that are beginning to employ machine learning and AI in healthcare applications, Professor Antoniades notes that the more data that we can gather from patients now, the better the ability of the FAI technology to predict heart attacks will be in the future.

‘The key to improving the diagnostic ability of these technologies is to include data from multiple cohorts in multiple countries,’ he says. ‘The more data you can put in, and the wider the pool it’s collected from, the better the computer will be at discerning what is and what isn’t a sign of future health risk.’

The researchers featured in this article would like to acknowledge their funders:

Charalambos Antoniades: British Heart Foundation, NIHR Oxford BRC

David Clifton: EPSRC, Wellcome Trust, NIHR, NERC, DfID, UNICEF

Alison Noble: EPSRC, InnovateUK, ERC, GCRF