British researchers have developed a form of artificial intelligence (AI) that is able to diagnose certain medical conditions based on scans, something that it is hoped could be widely rolled out to save the NHS billions of pounds.
The team at John Radcliffe Hospital in Oxford wanted to work on heart scans to improve their efficacy. Currently, cardiologists use the timing of heart beats in scans to see if there is a problem with a patient's heart, but this can be vulnerable to errors.
Indeed, although 60,000 heart scans are carried out each year, 12,000 are misdiagnosed by medical professionals. As a result, patients often go home with an all-clear but eventually have a heart attack, or they are subjected to unnecessary operations that cost the NHS an estimated £600 million.
Instead of relying on doctors to interpret the scans, the researchers developed an AI system called Ultronics, which was trained by being fed the scans from 1,000 previous patients and data on whether or not they went on to have a heart attack.
As a result, Ultronics was able to analyse risk from new scans and give a positive reading if it was likely that a person would have future heart problems. It has now been tested in clinical trials in six cardiology units and the results are to be published in a medical journal later this year.
Professor Paul Leeson, a cardiologist who developed the system, said: "As cardiologists, we accept that we don't always get it right at the moment. But now there is a possibility that way may be able to do better."
Sir John Bell told BBC News: "There is about £2.2 billion spent on pathology services in the NHS. You may be able to reduce that by 50 per cent. AI may be the thing that saves the NHS."
If it does prove to be a success, Ultronics could be rolled out to NHS hospitals across the country as early as this summer. In future, it may even be that AI is available in pharmacies and other community healthcare providers to boost services available to the public.
Meanwhile, another AI system is being tested to check for signs of lung cancer by looking for clumps of malignant cells called nodules. It is hoped that 4,000 people a year could be diagnosed with the disease much earlier than is currently the case, potentially improving their survival rates significantly.