We talked to Max Little, MIT mathematician, TED Speaker, and the man behind the science that can predict Parkinson’s from voice recordings with 99% accuracy.
Andy Grove was born to a middle-class Jewish family in Budapest in 1936. When he was eight, the Nazis occupied Hungary and deported nearly 500,000 Jews to concentration camps, including Auschwitz. For 12 years Andy and his family faced persecution, until in 1956, during the Hungarian Revolution, at the age of 20, Andy left his home and family and escaped across the border into Austria.

Penniless and barely able to speak English he eventually made his way to the United States where he earned a bachelor’s degree in chemical engineering from the City College of New York in 1960 before completing a PhD at the University of California, Berkeley.

Shortly after graduating, he joined Intel on the day of its incorporation. He was the fourth employee, director of engineering. Twenty years later he would be Chairman and CEO of Intel, taking the company from a revenue of $2,672 in its first year to $20.8 billion in 1997.

Intel’s CEO and Parkinson’s

 

In 2000, at the age of 64, Andy Grove was diagnosed with Parkinson’s disease. Over time he began to have trouble falling asleep and felt constantly lethargic. Then he started to experience hand trembling and found it increasingly difficult to stand and to walk because everything felt like it was moving in slow motion. 

As his condition worsened, and he accepted that there was no cure for his debilitation, Andy invested some of his own money into creating an Intel research team, which captured over 7,000 voice recordings in a clinical trial. The samples were a mixture of voices of people with Parkinson’s and voices of people without.

They hoped to build a device capable of testing for early signs of Parkinson’s. He hoped this would give others in the future a better chance to prepare for what was to come when they had early-stage Parkinson’s. The trouble was the Intel researchers had no idea what to do with the data once they had it.

Intel researchers meet Max Little

 

At the same time Max Little, a Birmingham mathematician and software developer, was at Oxford working on a PhD in clinical voice analysis. Max recalls,  “I had a chance encounter with a researcher at Intel Corporation. One of the co-founders of Intel was diagnosed with Parkinson’s and, realising that there are no objective methods that could be used to score the symptoms over time, he put some of his own money into Intel to develop a testing device that could do just that”.

When the Intel researcher heard Max was doing work on voice analysis he asked if he would take a look at their voice recordings. Max agreed to take a look at the data and over 6 months he developed an algorithm that could predict Parkinson’s disease from voice files with 86% accuracy.


Normal voice


Voice with Parkinson’s

 

To get a gauge of the voice files Max analysed, here is a sample of a normal voice compared to a voice with Parkinson’s. Source: MedGadget


Max leads world’s largest voice-based trial in Parkinson’s

 

Later Max went on to run a much larger trial involving 17,000 participants. This is the largest voice-based trial in Parkinson’s to-date. He recalls, “I wanted to test whether it was possible to get good enough voice data through the telephone in order to replicate my earlier research. We had an overwhelming response.

In lab-based tests, we were able to show that we can discriminate Parkinson’s from controls with high accuracy (close to 99%) from voice recordings collected over the telephone

 

This said, Max is careful not to oversimplify what the his team can achieve, explaining “applying this outside the clinic is another prospect altogether and we probably cannot achieve that high level of accuracy in such uncontrolled conditions”.

Parkinson’s Voice Initiative

 

Today, as well as being a Professor of Mathematics at Aston University, UK, Max is Director of the Parkinson’s Voice Initiative, which aims to show that the work they have done in the lab can be applied to the real world using telephone clips.

The benefits of achieving that goal are myriad. For one thing, voice tests for Parkinson’s disease would help patients to get a diagnosis. 


Max Little’s 2012 TED Talk, watched by more than 1 million people. Source: TED


In addition to working with telephone recordings Max and his team at Parkinson’s Voice Initiative are also currently working with smartwatch technology to detect early onset of Parkinson’s. When we spoke to him earlier this month he said, his team “have several projects under development. One is researching the use of smartphone-based testing to detect precursor symptoms of Parkinson’s in people with a sleep disorder known to precede diagnosis in many cases”.

Early results show the devices can predict Parkinson’s with 85-90% accuracy. This may lead to a cheap way of detecting the disorder in those at risk. Max also shared,

We are also trialing a method with a smartwatch that collects data from walking and other movement impairments so patients have an objective, accurate record of their symptoms when they visit their doctor

 

“Today we work with clinical teams around the world, where data from wearables/smartphones is captured in clinical trials and sent to us remotely. Then we work on developing mathematical models of the data, so that we can extract specific clinically-useful information from it.

“A lot of this is about finding ways to summarise this very large amount of data into something that is more meaningful to the clinician or patient. To do this, particularly for wearable devices, we need to know a lot about the context under which it was collected, the behaviour of the patient and the relationship between these and the underlying disease process”.

Max is hopeful but cautious not to sell big promises. “It can be hard to make general statements about how we solve these kinds of problems, because every clinical analysis problem is unique” he explained. And when I asked him what lesson he’d learned from all this research he said the biggest lesson was “how much variation there is in common disease areas. Every patient is unique and that makes the job of trying to analyse data extremely difficult”.

Predictions about the future of AI in healthcare

 

While I had Max I thought I’d ask him what predictions he had for the future of AI in healthcare. He shared, “Over the years, as I have become more experienced with AI and the evidence has started to accumulate, I’ve become more confident that the pending AI revolution in healthcare is largely a chimera.

“Much of the hype is unlikely to come true anytime soon. There may be some straightforward, routine tasks, like clinical operations or hospital systems that might be automated, but I don’t think that the job of doctors, or any clinical staff, is actually under threat.

“Basically, the reason is that, despite impressive progress in mathematical AI algorithms over the years, these are only applicable to simple problems where the conditions can be very carefully controlled. In the real, messy world of practical medicine applied to the complexity of biology and human life, almost none of these conditions hold true.

“In reality then, state-of-the-art AI doesn’t really work in practice, despite what you might read in the published literature or the popular press. It will take a lot longer and some major breakthroughs, I think”, Max concludes.

Contribute to the Parkinson’s Voice Initiative

 

The Parkinson’s Voice Initiative plans to perform additional research in the future. If you would like to be involved in a trial, register your interest at www.parkinsonsvoice.org.

Max Little

Project Director, Parkinson’s Voice Initiative

Max Little is a mathematician and Associate Professor at Aston University in the UK. He holds a degree in mathematics from the University of Oxford and is a Wellcome Trust/MIT fellow. Max is a TEDGlobal 2012 Fellow with two TED talks under his belt. He is also Project Director of Parkinson’s Voice Initiative, which develops methods for detecting Parkinson's disease from voice recordings.

Read More