How artificial intelligence, and a cup of tea, could help diagnose Alzheimer’s disease
by Guest Author on 26 Oct 2018
Congratulations to MRC PhD student Natasha Clarke, from St George’s, University of London, winner of our 2018 Max Perutz Science Writing Award. In her award-winning article she describes how teaching machines to detect changes in language could help with early diagnosis of Alzheimer’s disease.
I’d like to give you a quick task. How do you make a cup of tea? Describe it out loud. Whilst this could lead to some controversies (milk in first, or last?) it seems fairly simple. But what if I told you that this task could help diagnose Alzheimer’s disease?
A person receives a diagnosis of dementia, of which Alzheimer’s is the leading cause, every three minutes in the UK – that’s one in the time it takes a kettle to boil. In Alzheimer’s the hippocampus – the part of the brain we need to form and hold on to memories – starts to shrivel. Abnormal proteins build up in and around brain cells, killing them. Memories start to fade. And so does language.
This process is a slow one. We now know that by the time a person goes to their doctor with concerns about their memory, Alzheimer’s could have been lying undetected for 10 years. Complex tests, brain scans and lumbar punctures can give a good indication if someone has Alzheimer’s at this stage, but it’s too late. How can we bring the diagnosis forward?
Brain scans are getting more advanced, but they’re expensive. We need something that’s sensitive to change, cheap and readily available – like a blood test. Something that will help us detect who is in the earliest stages of Alzheimer’s, give medications sooner, and stop the disease in its tracks.
Language. There’s more information contained in language than we could ever imagine. The words we use can reveal our gender, how much power we hold, whether we’re angry, sad or happy. It’s a mirror to our inner selves, and very difficult to mask. And, language can also reveal signs of Alzheimer’s.
Ground-breaking studies have found changes in what people say, or how they say it, years before they’re diagnosed. President Ronald Reagan was diagnosed with Alzheimer’s after
he left office, but showed signs that his brain was already changing in speeches made six years earlier.
Researchers could tell from essays written by nuns which of them would go on to develop Alzheimer’s. Fans and critics alike panned Iris Murdoch’s final book, which was later found to have used language very differently from novels written earlier in her career. She was diagnosed with Alzheimer’s at 76.
This is what my research is doing. Finding clues in language that show the brain is changing. Building a method for detecting Alzheimer’s before the memories start to fade, and the disease starts to spread.
The vast majority of us aren’t presidents, nuns, or famous authors, though. We need to study the language of ‘everyday people’. So, I’m analysing the language of volunteers diagnosed with early Alzheimer’s disease or Mild Cognitive Impairment (MCI), where people experience some of the symptoms of Alzheimer’s and are at a higher risk of developing it later.
I ask these volunteers to name as many animals as they can in one minute, to describe a picture, tell me a story and to describe how they make a cup of tea. I record their speech, and then search through the hundreds of features hidden in their language, and over one year follow them up to see how this changes.
But searching through these hundreds of features is like looking for a needle in a hay stack. This is where Artificial Intelligence comes in. AI relies on Machine Learning, where a computer is given lots of data and learns something useful.
Take your spam email filter: it’s been trained to recognise spam compared to real emails, by searching through the characteristics, or features, of each. Perhaps spam emails contain more website links? Or real emails tend to be longer? Once it’s seen enough emails to learn what is different about the two groups, when you get a new email it knows what to look out for, and can banish that spam to the right folder.
This is how I will compare the language of my volunteers with Alzheimer’s and MCI to the language used by people ageing healthily, and find the needle in the haystack of spoken words. I will train a computer to learn which features are important for identifying disease, so we know what signs to look for. For example, I’m looking at how common the words used by each group are. Is speech affected by Alzheimer’s more predictable?
Blood tests look for biological markers of disease. I’m looking for a marker in language. Just like a blood test, it’s sensitive to change, cheap and readily available, and could reveal Alzheimer’s years before other symptoms start to show. With the help of AI, I hope that in the future diagnosing Alzheimer’s will be quick, inexpensive and painless – as simple as making a cup of tea.
Read the 2017 winning article.
Find out more about the competition.