Say 'Ahhh': A Simpler Way To Detect Parkinson's
There's currently no cure for Parkinson's, a debilitating neurological disease. There's also no blood test that can detect it, meaning early intervention is almost impossible.
But soon there might be a shockingly easy way to screen for Parkinson's disease. It would be as simple as picking up the telephone and saying "ahhh."
"There's some evidence, admittedly weak, that voice disturbances may well be one of the first or early indicator of the disease," mathematician Max Little tells weekends on All Things Considered host Guy Raz.
Little is head of the Parkinson's Voice Initiative and he's created an algorithm that can determine whether or not a person has Parkinson's just by the sound of their voice.
Right now, the algorithm has a 99 percent success rate.
A Surprising Result
Little worked on this algorithm while he was getting his PhD at Oxford. It didn't occur to him that it could be used to detect Parkinson's disease until a chance encounter with a researcher from Intel.
"He came up to me and said, 'Max, we've got vast amounts of data recorded from Parkinson's disease patients and we really don't know what to do with it. So perhaps your algorithms might be useful for analyzing this,'" Little remembers.
The recordings were voices of people both with and without Parkinson's and Little's algorithm sorted them. It was 86 percent accurate.
"And that's when we thought 'Ahh, okay, we got something here,'" Little says.
Now, he's looking to crowdsource more voice recordings of people with and without Parkinson's to make the algorithm even more accurate.
Last month, Little and his team asked people all over the world to call up their phone banks and record their voices. The caller follows a series of prompts — asking them to say "ahh" for as long as they can and say a few sentences.
"We expected to it to run for six months and collect 10,000 voice recordings over that period," Little says. "In just under a month, we got 5,000 voice recordings."
The algorithm listens for three main clusters of symptoms in the voice: vocal fold tremors, breathiness and weakness, and the way the jaw, tongue and lips fluctuate during speech.
Little hopes the algorithm will improve enough so that eventually a person could get a diagnosis and track their illness through the phone.
He admits it's still in the early stages, but it could be the simplest way to diagnosis Parkinson's — and a way to help patients get earlier treatment.
"It's a very tantalizing possibility. If voice is one of the early indicators of the disease, then we have a very cheap way for looking for those who are susceptible in the population," Little says.
GUY RAZ, HOST:
Parkinson's is one of the hardest neurological diseases to treat. There's no cure and it's very difficult to diagnose, particularly in its early stages. Part of the reason is that blood tests don't reveal whether you have it or not. Well, that may be about to change in a shockingly easy way, and it could start by placing a phone call and saying, ahhh.
UNIDENTIFIED MAN #1: Hello. Welcome to the Parkinson's Voice Initiative. We record your voice for clinical research.
MAX LITTLE: We play someone various prompts that tells them exactly what to do. In particular, the primary thing that we're looking for is for them to just say, ahhh, for as long as they can.
RAZ: That's Max Little, head of the Parkinson's Voice Initiative. He has created an algorithm that could determine whether or not a person has Parkinson's just by the sound of their voice. Max Little is an applied mathematician. And a few years ago, a researcher from Intel Corporation, which pours millions of dollars into Parkinson's research, approached him, and here's what happened next.
LITTLE: They asked me, can your algorithm separate out those who have Parkinson's from those who don't? And so we tried it out, and we were able to separate healthy controls from Parkinson's disease patients to 86 percent accuracy, and that's when we thought, ah, OK, we've got something here.
RAZ: How accurate is the algorithm now?
LITTLE: So we know that we can detect whether somebody has the disease or not to 99 percent accuracy.
RAZ: Ninety-nine percent.
LITTLE: That's right. Yeah.
RAZ: So tell me how does your algorithm work.
LITTLE: There are three sort of cluster of symptoms that we're looking at. The first group are to do with vocal fold tremors. So this is looking for a tremor in the way that your vowel sounds are produced, for example. The second type of cluster of indicators that we're looking for are to do with breathiness and weakness. The third kind of indicator that we're looking for - and this is relatively new actually in the literature. It's a new indicator.
And that has to do with the way in which your jaw or your lips, your tongue will slightly fluctuate during speech. And that turns out to be a very important indicator as well, but that's a very recent finding, actually.
RAZ: So, Max, you have decided to create a database of voices - people with and without Parkinson's - in order to, presumably, to improve the rate at which you can figure out whether a person has Parkinson's based on their voice. Let's listen to one of those voices. This is a person without Parkinson's, and they're just saying, ahhh.
UNIDENTIFIED WOMAN: Ahhh.
RAZ: OK. Let's now hear somebody who has Parkinson's doing the same thing.
UNIDENTIFIED MAN #2: Ahhh.
RAZ: Oh, yeah. You can definitely hear that difference.
RAZ: Explain what we're hearing here.
LITTLE: So this is really interesting. This person has quite obvious vocal fold tremor. It's the first of the three clusters of indicators that I pointed out. There is also an element of weakness, certain out of breathness in their voice. That's also typical. So you can see two classical indicators of these symptoms occurring in this case.
RAZ: Do you think there's any link between diagnosing Parkinson's through this voice technology and getting closer to curing it?
LITTLE: Well, this, of course, is very exciting. There's some evidence, admittedly weak, that voice disturbances may well be one of the first or early indicators of the disease. Now, if that is true, that means that if we can actually get to the population and record a large number of voices, we might have the opportunity to detect those people who could benefit from neuroprotective treatments. That is basically to look for cures.
Because, of course, the problem with Parkinson's disease is that by the time it's being diagnosed, the damage is irreparable. So if voice - and I say it's a big if because we don't know this for sure, but it's a very tantalizing possibility - if voice is one the early indicators of the disease, then we have a very cheap way of looking for those who are susceptible in the population, then we can target neuroprotective treatment to them to see whether they respond or not.
RAZ: That's Max Little. He's a post-doctoral research fellow at MIT and the head of the Parkinson's Voice Initiative. If you want to find out how you can add your voice to the project, go to www.parkinsonsvoice.org. Max Little, thanks.
LITTLE: You're welcome. Transcript provided by NPR, Copyright NPR.