Technology that uses artificial intelligence (AI), data analytics and 3D imaging to perform non-invasive cardiac tests has been deployed in 28 NHS hospitals over the past six months.
The technology, developed by US medical technology (medtech) firm HeartFlow, is one of seven innovative treatments for a range of medical conditions supported by funding from the government’s Accelerated Access Partnership Rapid Uptake Products programme.
According to an evaluation by the National Institute for Health and Care Excellence (Nice), use of this technology could save the NHS up to £214 per patient, equivalent to £9m a year, when diagnosing coronary heart disease, which is estimated to cause 66,000 deaths each year in the UK.
Diagnosing heart conditions has traditionally been an invasive process, often involving the insertion of a catheter into a patient’s artery through either their groin or an arm, but the HeartFlow technology can produce the same results, and in a much shorter timeframe, with just a CT scan.
“The way we use it clinically is I see a patient, I scan a patient, and that takes about 10 to 15 minutes. I then have a look at the images we’ve taken from the CT scan and if I think we need more functional information, I can send it straight to HeartFlow,” said Tim Fairbairn, clinical lead at Liverpool Heart and Chest Hospital, a user of the product.
“As soon as it hits their [HeartFlow’s] system, it comes back with a turnaround time of four to five hours. It doesn’t matter what time of day it is, they just come back with a report, which is then integrated into our viewing system.”
The HeartFlow report can be compared with 3D computer models of the heart’s blood vessels and original patient data to determine the cause of the problem, said Fairbairn.
“It means [the patient] can have a more accurate answer within five hours, whereas traditionally I might have needed to get another test done, which I would then have to request and book,” he said. “It’s probably saving six to eight weeks in terms of timeframe for the patient and doctor to get that accurate assessment.”
At HeartFlow’s end, the data is analysed by AI technology after being uploaded to the company’s cloud, where the results are double-checked by a human analyst in one of two US-based production facilities.
“You need a lot of data to do this, but it has to be quality curated and annotated data,” said HeartFlow CEO Charles Taylor. “We have large amounts of data coming in from all over the world through the cloud, which then uses automated algorithms to process it before we have a person look at the result of the model in context with the image data, and if it doesn’t match, we fix it.
“Although it’s a new technology and not everybody fully understands certain aspects, compared with other technologies it certainly does seem to improve patient care”
Tim Fairbairn, Liverpool Heart and Chest Hospital
“Once you’ve done that thousands of times, the annotated model, with the deep learning method, learns how to do something a person does, but it learns how to do something perfectly reproducible – it learns how to read through areas of disease a person can’t.”
HeartFlow’s current algorithm is trained using data from almost 10,000 patients, all of which has been de-identified and is stored locally by the hospital or practice rather than on HeartFlow servers.
“There is still a black box aspect to it, whereby clinicians don’t understand exactly what is done, but the key bit is trusting the data,” said Fairbairn. “Although it’s a new technology and not everybody fully understands certain aspects, compared with other technologies it certainly does seem to improve patient care.”
HeartFlow has plans to expand to 50 NHS hospitals by 2019, with the help of funding from NHS England’s Innovation and Technology Payment programme.
Taylor, originally an engineer who became interested in computer modelling, first came up with the idea for the technology after noticing that doctors worked on a trial and error basis when diagnosing heart conditions.
“Why do we have to do it that way? Why can’t we create a computer model of somebody, of their arteries, and use it to better diagnose and understand the condition they have – use it to actually map out their disease and treatment?” he said.
“So I had the idea of combining computer analysis together with imaging, and to put those together to contemplate doing things like modelling blood flow in individual cases rather than idealised computer models created for an average person.”
