What actually is AI? How does it work? And should we be frightened of it?
We hear the words a lot — you can scarcely pick up a paper or read a news site without coming across some mention of artificial intelligence. Once the content of futuristic science fiction, today artificial intelligence is being employed in diverse ways that often claim to enrich and enhance our lives — from asking Alexa to arrange a furniture collection, self-driving cars and improving diagnosis in health care. But we still often associate machine learning with scary headlines. So are we worrying unnecessarily? Or should we really beware the rise of the sentient robot?
What is ‘Big Data’?
Big data is a term given to sets of data that are so large or so complex that they aren’t able to be analysed using traditional methods of data processing, eg someone having to manually sift through it.
“Big data is for machines, small data is for people”.
Rapid computer analysis of large amounts of information has created a landscape that makes it possible to tackle some of the biggest questions in science, which would previously have been impossible. In recent years, increases in computing power have enabled software to be developed for doctors to use.
BHF Professor Sir Rory Collins is the Principle Investigator and Chief Executive of UK BioBank, a major national and international health resource, which has collected information, health data and samples from 500,000 people in the UK. This huge amount of information, part-funded by the BHF, is now being used by research teams across the globe to investigate a variety of serious and life threatening illnesses, including heart disease. UK Biobank is a powerful tool to help research teams understand more about how our genetics and environments contribute to different life-threatening diseases.
Big data projects using data like that created by UK Biobank have huge potential. In 2018 an international team used this data to create a computer programme that could predict our risk of future coronary heart disease, even in children. Coronary heart disease (CHD) is when the arteries that supply blood to your heart muscle become narrowed by a gradual build-up of fatty material within their walls
At the moment doctors predict our risk of developing CHD by looking at a number of factors — our family history, whether we smoke, our diet, and whether we have other risk factors like high blood pressure or high cholesterol. However, many of these factors change as we age, meaning that those most at risk may not know until later in their life.
The Genomic Risk Score (GRS) instead looks at our genes, which don’t change as we age — meaning this test can be used to tell us our risk much earlier in life. The BHF-funded team, including BHF Medical Director Professor Sir Nilesh Samani, analysed the genes of the 500,000 people who are part of the UK biobank, 22,000 of whom had coronary heart disease to build the GRS tool. It looks at 1.7 million variations in a person’s DNA, and then calculates their underlying genetic risk for CHD.
The team showed that the GRS was better at predicting someone’s risk of developing CHD than by looking at their observable risk factors. In fact, the GRS’ ability to predict CHD was actually largely independent of blood pressure and cholesterol level, suggesting that our genes don’t increase our risk by just targeting those things — it is much more complex than that.
This observation could also help to explain why people with healthy lifestyles and no conventional risk factors can still be struck by a devastating heart attack.
By establishing someone’s risk of developing CHD earlier in life, this test allows us to give them better, more personalized advice — whilst we can’t change our genes, we can make lifestyle decisions based on our known risks. The earlier we can do this, the better.
What is AI?
Artificial intelligence (AI) is the term used when a computer mimics human intelligence. What separates AI from other types of computer programs is that they can ‘learn’ and improve at tasks.
Machine learning is a type of artificial intelligence that uses algorithms (a set of rules that a computer uses to make a calculation) to look for patterns in data and then makes decisions based on these patterns with limited, if any, input from humans. It looks for patterns in many different types of data — from scrutinising images to analysing health data.
Deep learning is another type of artificial intelligence — unsupervised machine learning. Standard machine learning needs a person to program the computer to tell in what patterns or features it should be looking for when analysing the data. With deep learning, the human input is much smaller, using multiple interconnected layers of analysis (called neural networks) the computers can analyse things more similarly to the way a human brain does.
Is AI smarter than humans?
This is complex because you can’t think of intelligence as a straight line with ‘smart’ at one end and ‘not smart’ at the other.
In some areas you might think that computers are already smarter than us — they can calculate things more quickly, make fewer assumptions when processing data and store and search a lot more data than we can. But their emotional intelligence and creativity has not yet matched us and, according to experts working in the field, isn’t likely to reach our level for a while.
Health data is also very complex and AI can’t always tell us if something caused something else (causal), only that one thing is linked to another (correlation).
What could big data and AI really mean for our heart health?
AI has already been able to identify early signs of heart failure, and it is predicted to be used in the future across the NHS to help diagnose and treat a variety of heart and circulatory diseases.
Dr Declan O’Regan and his team, based at Imperial College London, are using artificial intelligence (AI) to interpret heart scans and tests to try to predict how, why and when people with certain heart conditions will go on to develop heart failure.
Heart failure means that the heart is not pumping blood around our body as effectively as it should. This can happen after a heart attack, as the heart muscle can become irreversibly damaged. As more people are now surviving heart attacks, thanks to research, the number of people living with heart failure is increasing.
Currently, there is no cure available for those living with heart failure, but they can be given treatments to help manage their symptoms. Doctors therefore need to know when people in the early stages of heart failure are likely to get worse, so they can provide the best treatment. But this is not always straightforward as current measurements don’t allow us to accurately predict how their condition will progress.
Dr O’Regan’s team is using AI to try to accurately predict what will happen, so that people can receive the best treatment as soon as possible. They are using AI to interpret hundreds of heart scans and build a detailed three-dimensional model of a person’s heart, before training the computer to recognise the earliest signs of heart failure.
They want to compare how accurately the computer predicts the likely outcomes for people with serious conditions that can lead to heart failure. If this work can reveal a better way to predict how heart failure progresses, it could lay the foundations for introducing AI into how these people are cared for and could transform the way scans and other tests are used to make the best decisions about a person’s treatment.
What are the challenges for AI in healthcare?
Often, using AI in a medical setting requires the use of large amounts of patient data. The rules outlining the use and sharing of this data are, rightly, very strict. But this can also make it much more difficult for research teams to access the amounts of data they need for their research. With the involvement of the tech sector in this kind of research, care has to be taken to make sure that people are aware of exactly how their data would be used and the possible risks associated with this, no matter how small those risks may be. In 2017 the National Data Guardian, Dame Fiona Caldicott, published a report which outlined the importance of public trust in data use — that people should have choice about the use of their health data, and that there should be a clear dialogue about how and why their data us used.
Which leads us to another challenge we need to overcome before the use of AI in health research becomes more widespread: how it is perceived. In a recent BHF survey of people with heart and circulatory disease, 9% of people felt very confident in the security of anonymized personal health data being used in artificial technologies. The field still clearly has a long way to go to help people feel more confident in sharing their data for research purposes.
Additionally, all current machine learning requires some form of initial input and supervision from human researchers. This means that human biases can sometimes creep in to the algorithms the machines use to make their decisions. Research teams therefore need to work hard to prevent these biases leaking into the algorithms and perpetuating the problem.
The way data is stored and collected can also be a challenge for the field — different teams, hospitals and agencies record and store the information in different ways, making it really difficult to compare different sets of data and limiting the way the two sets of information can be used effectively. Therefore, a big focus for research teams in this area is on data standardisation — so all information can be used, compared and studied effectively.
What next for AI in heart and circulatory disease research?
The BHF are committed to accelerating progress in data science to find new ways to better diagnose and treat the millions of people across the UK living with heart and circulatory diseases. To help drive more of this kind of research, since 2017 the BHF has partnered with the Turing Institute to fund research in this area — bringing together data scientists and cardiovascular researchers to find solutions to important health problems. In the first round of this partnership Dr Angela Wood, Senior University Lecturer in Biostatistics at The University of Cambridge, has been awarded funding to develop a machine learning tool to predict which people are likely to suffer a life-threatening heart attack or stroke.
“It’s only recently that we’ve had the technology to process the huge amount of data available in health records and use it to our own advantage. New algorithms could allow us to pick up entirely new and detailed patterns in people’s past health to predict their risk of future events — ultimately saving lives.”
In addition the BHF is working in partnership with Health Data Research UK (HDR UK) in to establish the infrastructure and expertise needed to propel the UK to become a world leader in data science by providing access to a wide range of data health resources needed to help research teams better understand all kinds of heart and circulatory disease.
Whilst there are certainly challenges in effectively and securely bringing AI into everyday use in healthcare, the opportunities the technology could bring to support those living with heart and circulatory conditions are definitely something to be excited about. From using AI to help determine when someone might need treatment, to helping to analyse clinical data and decide the best treatment for us on a person by person basis, the possible impact on our health could be huge.
If you liked this, you can find follow the BHF publication on Medium and check out more about BHF research here:
- Broken Bones and Broken Hearts: How are they connected?
- Teeny tiny solutions to one of the world’s biggest heart aches — Honey, I shrunk the science
- Digital medicine: ‘the algorithm will see you now’
What is artificial intelligence and what could it mean for our heart health? was originally published in British Heart Foundation on Medium, where people are continuing the conversation by highlighting and responding to this story.