Save lives, and save money for the health service. It’s an outcome few could complain about, and it’s happening because of artificial intelligence (AI).
The medical diagnostics company behind this development, Ultromics, was spun out from Oxford University research around 18 months ago. Using the power of AI, Ultromics aims to improve the accuracy of echocardiogram interpretation to above 90% – substantially better than the 80% currently achieved by human doctors.
This, say the company’s founders, will save lives by identifying more people at risk of heart disease and – by reducing the number of patients unnecessarily sent to theatre – potentially save billions for health services around the world.
... you need an expert to interpret the images and reach a diagnosis. When the expert is good, then the test can be very accurate. But because levels of experience vary, this can be difficult to control.
Paul Leeson is Professor of Cardiovascular Medicine in Oxford’s Radcliffe Department of Medicine and one of the founders of Ultromics. He says: ‘Echocardiography is the most widely used imaging test in people with heart disease. In most hospitals, over ten times more echocardiograms are performed than any other imaging test in cardiology. This is because echocardiograms can be performed quickly, anywhere in the hospital, including at the bedside or in the clinic. Echocardiograms are also performed in the community and in remote locations, or areas where resources are limited.
‘However, you need an expert to interpret the images and reach a diagnosis. When the expert is good, then the test can be very accurate. But because levels of experience vary, this can be difficult to control.
‘We wanted to fix this by using AI methods to standardise how images are analysed, lifting the quality of interpretation so that it is always as good, or better, than an expert reader. To do this, we built up databases of hundreds of thousands of echocardiography images linked to information about what was unique about the person who was being imaged and what happened to them over time. By combining machine learning with clinical know-how, we were able to identify associations between features hidden within the echocardiography images and what happens to patients. Doctors can then use this information to decide how to look after the patient.’
Ultromics’ co-founder and CEO Ross Upton is, perhaps unusually in a University spinout company, a current graduate student at Oxford, nearing completion of his DPhil in cardiovascular medicine under Professor Leeson. Upton had the idea of applying AI and machine learning techniques to this field after learning of the shortfall in the accuracy of diagnosis. Within two years, Ultromics had been spun out of the University with the help of Oxford University Innovation – Oxford’s research commercialisation arm – attracting more than £10m in investment led by the Oxford Sciences Innovation fund.
Upton says: ‘The first product of Ultromics, EchoGo, is based on extracting features from stress echo images and using a supervised machine learning model to predict the outcome of a patient one year following the test. The features we extract from the images are all biologically relevant to the disease process – some of which are clinically known and others which are entirely novel features that we have patented.
By using the AI technology to ensure consistent and accurate interpretation, you can reduce the need for unnecessary additional investigations and ensure you do not miss disease. This improves the care of the patient and significantly reduces costs for the NHS
‘We used one-year patient outcomes as the gold standard, rather than how someone has reported the scan, because we know operators interpret the scan correctly only 80% of the time. We therefore need to follow up the research participants for a year after the exam to see what actually happened to them after the test. If the test is interpreted incorrectly, the patient would get sent for an angiogram unnecessarily; if the test was reported as normal but there was underlying disease, then the patient would get sent home when they should have been sent for an angiogram. It’s these errors that EchoGo is going to reduce.’
Professor Leeson adds: ‘Stress echocardiography is used widely across the world – it is the most commonly used functional imaging test for coronary artery disease in the UK. By using the AI technology to ensure consistent and accurate interpretation, you can reduce the need for unnecessary additional investigations and ensure you do not miss disease. This improves the care of the patient and significantly reduces costs for the NHS. Also, because stress echocardiography uses ultrasound equipment that is already available in hospitals and can be delivered by existing clinical staff, it means hospitals can more carefully consider whether they need to spend money on expensive new tests and infrastructure or instead put their existing infrastructure to better use.’
The next step for the company, says Upton, is to achieve a CE mark and clearance from the US Food and Drug Administration – hopefully by the spring of next year, so that EchoGo can be introduced to clinics and begin improving patient outcomes. He adds: ‘We are also looking to expand our already large-scale clinical trial to 30 different hospitals across the NHS. The next innovation is to completely automate EchoGo, which will help provide an instantaneous result to clinicians. This will be done by utilising newer deep learning frameworks, which are being refined at the moment by our research and development team. Following that, we will look to tackle other disease areas within echocardiography, such as heart failure and valve disease.’
Reflecting on the process of spinning out a commercial company from University research, Professor Leeson says: ‘A lot of companies are spun out from Oxford, but that is not because it is an easy thing to do. The number reflects the amount of high-quality, truly “translatable” research being carried out by investigators in departments. This is coupled with very effective and experienced support from Oxford University Innovation. From concept to spinout took us two years, and we had to get over a range of hurdles on the way, including securing IP and patents, being awarded pre-spinout seed funding to build aspects of the technology that would be attractive to investors, and, finally, convincing a lead investor to invest in both the technology and us, as founders.
‘You have to have a really game-changing idea, with science to back it up, to convince investors. Even at that stage, negotiating the details of how the company is formed and its ongoing relationship with the University can take several months to arrange. After that, the Oxford environment, with supportive backers such as Oxford Sciences Innovation, means the acceleration and growth of the company can be very rapid.’