AI vs heart disease: How machine learning could help doctors tackle heart problems before they happen

A system that spun out of Stanford is using AI and machine learning to help doctors visualise patients' arteries and spare them invasive tests.
Written by Jo Best, Contributor

"The NHS is traditionally like an oil tanker – it moves slowly. For 39 separate units to ask for this, this is something that isn't at the beginning any more, it's now mainstream."

Dr Derek Connolly, a consultant interventional cardiologist at the Sandwell and West Birmingham Hospitals NHS Trust, has been using HeartFlow's technology for around a year. Over the past 12 months, he's seen interest in using the tech, which helps doctors see inside the coronary arteries, grow in the UK.

HeartFlow, a medtech company based in California, recently received $500m in funding to help it commercialise its technology. Its algorithms analyse medical scans to produce 3D reconstructions of the coronary arteries, and makes use of machine learning and AI to improve accuracy.

The coronary arteries are the vessels that supply the heart with blood. If those vessels get narrowed – due to smoking or high cholesterol, for example – the heart doesn't get the blood it needs to keep on working effectively, and that can lead to a heart attack. Heart disease is the biggest killer of men in the UK, and the second biggest cause of mortality overall, accounting for over one in 10 deaths.

In order to bring down the number of people dying from heart disease, doctors would like to be able to catch patients before they arrive in A&E – an ounce of prevention really is worth a pound of cure. "By the time they've come to me, they've already had a heart attack. In many ways, what would be nicer would be if we could catch them before they've had a heart attack and change the future," Connolly said. 

And that's where HeartFlow comes in. The technology, which grew out of Stanford University research into fluid dynamics, takes data from CT scans of patients and builds a 3D representation of their hearts. That allows doctors to view exactly where, and how significantly, the coronary arteries are narrowed. 

SEE: How to implement AI and machine learning (ZDNet special report) | Download the report as a PDF (TechRepublic)

While it may sound simple, it marks a significant departure from previous practice – and one that could spare patients a lot of stress, waiting around, and unnecessary invasive tests. 

For any patient who turns up at the hospital with chest pain that's thought to be due to heart disease and then referred to the rapid access chest pain clinic, the first-line test is a cardiac CT. Those patients with evidence of seriously narrowed arteries may be offered an invasive angiogram – a procedure that sees them wheeled into the catheter laboratory and a wire inserted into a coronary artery, and the pressure measured at the top and bottom of the vessels. If the difference, known as fractional flow reserve (FFR), is under 80 percent, the patient will need stents – tubes put inside the arteries to keep them open. 

"Cardiac CT is really good, it's the most sensitive and specific test we've got, but it tends to exaggerate disease – the narrowing looks worse on cardiac CT than it does on invasive angiograms. The question is could we take that technology that we use in the cath lab and use it downstream?" said Connolly. 

By studying the flow of a material called contrast through patients' arteries during the CT scan of their heart, HeartFlow can calculate a patient's FFR without the need for the invasive angiography.

That's good news for the patient, says Connolly, who is spared the risks associated with the invasive testing. For example, the radiation a patient absorbs in a cardiac CT is a quarter of the amount they absorb in an invasive angiogram. "The more radiation you get, the more likely you are to induce a cancer, particularly in younger patients. The less radiation you use, the better. Secondly, [cardiac CT] saves on danger. For every diagnostic angiogram we do, one in 3,000 of the patients could have a serious illness or die. Because we're touching the coronary arteries, we can rip them and that's not unheard of. It's much safer to have a CT than to have an invasive angiogram."

SEE: Managing AI and ML in the enterprise (ZDNet special report) | Download the report as a PDF (TechRepublic)

By using HeartFlow to calculate the FFR, fewer patients end up having an invasive angiogram and the risks that go with it – and the ones that do go on to have an angiogram are the people that really need it. 

"For every five patients [that have a CT and HeartFlow], four patients go home knowing they don't need anything else. Half of those will be on cholesterol tablets, because they have early disease, half have normal coronary arteries. The one in five historically we send for a catheter angiogram, now seven out of 10 don't need to go to the cath lab for an invasive procedure, an operation, because the HeartFlow rules out that they have a significant stenosis," Connolly said.

Not only does it make life easier for patients, it makes life easier and cheaper for the hospitals – fewer invasive angiograms are done on those that don't need them, saving the money that would be spent on unnecessary procedures and freeing up the labs to only investigate those that are really at risk of having serious cardiovascular disease. "If we take someone to the cath lab, and they have normal coronaries, it takes up to an hour and that's a slot wasted. If we take someone where we think they're going to need a stent, that's useful time."

At the moment, the CT scans done in Birmingham are done and uploaded to HeartFlow's cloud. Once the company's algorithms have worked on the scans, it reports the results – for Birmingham patients, it's around three hours later. Connolly expects that delay to come down, perhaps to the point where a patient can have a scan, go for a coffee, and have the results by the time they've finished their drink. 

SEE: Sensor'd enterprise: IoT, ML, and big data (ZDNet special report) | Download the report as a PDF (TechRepublic)

While currently, the scans are done only on patients with cardiac chest pain, if the costs and time lag come down, the testing could be done on broader populations of people to diagnose the beginnings of heart disease and work to prevent it progressing. 

"The next question is, what if you take a high-risk population, say, patients that have diabetes, high blood pressure, high cholesterol, or they're heavy smokers, or all of the above – what percentage of those will have disease? The CT scanner might tell us. Then the question is, if they have narrowing [of their coronary arteries], would they benefit just from tablets, or bypass surgery, or stents? I think the future is going to be incredibly bright." 

While CT and HeartFlow can currently only show whether patients have significant narrowing in their arteries and should have stents put in, in future the system could give doctors an even greater steer on what to do next. Connolly predicts that before too long, HeartFlow will be able to suggest not only if stents are required, but also what size should be used, where it should be positioned, and what outcome might be expected, as its machine learning works on a greater and greater number of scans.

"At the moment, HeartFlow is good at predicting who doesn't need stents. Going forward, we should be able to predict what would happen if we put a stent in and that will help us make a clinical judgment as to whether a stent is right for the patient, or not."

The cardiologist rejects concerns that the growing use of AI in medicine could see technology de-skilling doctors, and instead views it as a way of giving physicians an additional evidence base to draw on. 

"Do I fear a machine telling me the probability to 17 decimal points the patient would be better with treatment x or y? Not in the slightest. We aren't there yet, but clearly technology has its uses and the era of the doctor just saying 'I think', with not much evidence, is going to disappear and we'll be able to plug people into the experience of hundreds of thousands or millions of patients."


How machine learning, drones, and robotics will transform the NHS and healthcare

For a health service in the middle of funding squeeze, can new technologies help to improve care and cut costs?

NHS and tech: Using digital transformation to rethink healthcare

Digital transformation can help the NHS make better use of scarce resources. That means rethinking some of the assumptions around healthcare.

Two cybersecurity myths you need to forget right now, if you want to stop the hackers

The wrong attitudes towards risk and complexity could leave you open to attack.

NHS and technology: Making the case for innovation

Physician, reboot thyself! The health service is caught between its creaky past and a shiny future. But change is needed, and fast.

Healthcare security nightmare: UK's NHS lost nearly 10K patient records last year (TechRepublic)

Despite having electronic record systems in place, 94% of NHS Trusts still use handwritten notes for patient record keeping, according to a report from Parliament Street.

Government websites hijacked by cryptocurrency-mining malware (CNET)

Over 4,000 websites worldwide were affected by the malware.

Editorial standards