Dial-A-Diagnosis? The Parkinson's Voice Initiative--In-Depth Doctor's Interview
Max Little, Post Doctoral Research Fellow at MIT Media Lab talks about using the voice as a biomarker for Parkinson’s Disease.
How did the Parkinson’s Voice Initiative start?
Dr. Little: We’ve been doing a lot of research on the extent to which voice could be used as a kind of ‘dry biomarker’ for Parkinson’s Disease. We were interested in the question of whether or not this technology could be deployed over the telephone. We’ve been doing a lot of lab studies: lab recording is important to test the idea but the problem is, of course, that these lab recordings will not be accessible to a large number of people. In previous research we’ve been able to show that we can get a very high accuracy detection rate using lab-based voice recordings. It would seem natural that the telephone could be one of the ways of deploying this technology. So that was the purpose of the Parkinson’s Voice Initiative: to test out whether we can make this work over the telephone.
What kind of program did it take to implement this kind of study?
Dr. Little: What’s required to implement this sort of research is an interactive voice recording service. We have a service in the UK that captures voice recordings using the standard telephony infrastructure that’s now very accessible. The service itself just captures the voice recordings but then we need to do the processing of these recordings afterwards; and this processing is fairly intensive. It requires a number of algorithms that are looking for specific features in the voice. It then requires machine learning algorithms that can make sense of those features.
So pretty complex stuff?
Dr. Little: It is quite complex. Certainly it’s using leading edge technologies -- for example computer science departments in universities around the world have been working on these kind of machine learning algorithms. We require those sorts of algorithms but we do have an advantage in the sense that the computing hardware is now becoming cheaper and faster so we can anticipate that in a few years time, the latest algorithms could run, for example on a smart phone in someone’s pocket.
So you could have it on your smart phone and just have doctors do diagnostic tests through their own smart phone?
Dr. Little: Yes, if that arrangement was valuable in a clinical practice of course.
You’ve done lab studies already, what were the results?
Dr. Little: In the lab studies we’ve done we’ve been able to show that we can get high detection accuracy – this is, accurately determining whether somebody has Parkinson’s or not. We are up to around 99% accuracy. This is now good enough to demonstrate screening capability. That’s the kind of technique that could be used in order to do triage, for example, to work out who might be at high risk of having the disease. The next stage is to pass that on to clinicians who could then do a more in-depth diagnosis, and as a result make appropriate treatment decisions.
What’s wrong with the test we have right now for Parkinson’s?
Dr. Little: Well, one of the problems with test that we have right now is that it requires you to be in the clinic. It’s going to be very difficult to get a large number of people to visit in order to screen for the signs of the disease, for example. When you have to go to the clinic you have all these associated logistical difficulties. So part of the motivation for developing this technology is about trying to bring these tests to a wider audience, and to reduce the cost involved in trying to do mass screening.
It can even help without lining drug timing and dosage?
Dr. Little: This is one clinical application that we think we can address with this sort of technology. If you have something as simple as a voice test then it’s feasible for you to take that test frequently during the daytime. And in this case what we can do is get accurate measurements of symptoms, so that you can have something like a day by day symptom diary showing how your symptoms are changing very regularly. Currently, the problem with measuring Parkinson’s is that because you have to go to the clinic, it’s very rare that people will actually get an objective measurement of their symptoms. Most of the time you have a consultation with a neurologist and it’s really all just anecdotal – they will ask you how you feel, that is, whether you feel that your symptoms are improving on this current treatment regime. A lot of that actually comes down to questions of timing and dosage. Having an objective symptom diary would really optimize that process, because it would allow clinicians to have an objective record of how someone’s symptoms have been changing over the day.
If they had the right dosage would they lose some of the tremor that you hear in their voice?
Dr. Little: We would expect that because tremor is one of the ‘canonical symptoms’. It is not the only symptom – there are many others which is why we’re looking at a lot of different features – but, yes, certainly that is what our research shows. So, as the drug wears off during the day then the symptoms come back and cease to be suppressed. This should also be evident in the voice.
You say you look for a hundred and thirty different things in the voice, what are some of them?
Dr. Little: All told now we’re looking for about a hundred and thirty different features. We’ve managed to reduce that down to an informative set of about ten different features. This set is what we would actually be using for practical computations in decision making. We have identified three different ‘classes’ of symptoms that we’re currently focusing on. The first class of symptoms has to do with vocal tremor – this is a ‘canonical symptom’ of Parkinson’s. The next cluster of symptoms have largely to do with breathiness, and what is perceived as weakness in the voice. The third cluster of symptoms are relatively new – we have identified these in our latest research. We’ve discovered that vocal tremor caused by involuntary movements in the lips and jaw seems indicative, and these are turning out to be quite a useful cluster of symptoms.
And these things that you’re looking for are evident from your first study?
Dr. Little: I should explain that what we’re doing is replicating the clinical diagnosis. So, someone goes to their neurologist and they get what a neurologist would consider as being a definitive diagnosis of Parkinson’s. Now, of course that doesn’t mean that there aren’t some people who are inaccurately diagnosed by that process, but we think it should be very, very small. What we’re able to do is to replicate that clinical diagnosis. What that says about the prevalence of voice symptoms in Parkinson’s may well conflict somewhat with what is known in with reference to less exhaustive studies of voice in Parkinson’s. Unfortunately, there are very few circumstances in which you have a clinical study which looks at voice in Parkinson’s as a primary outcome. They may well look at another primary outcome measure, e.g. contained in a scoring system called the Unified Parkinson’s Disease Rating Scale, and voice only accounts for a very small part of that. Unfortunately, most neurologists are not explicitly trained in voice analysis. Therefore, part of what we’re doing is trying to elaborate what is currently known about voice in Parkinson’s.
And the inaccuracy now is what thirty to forty percent in misdiagnoses in Parkinson’s? How can that help eliminate some of those as well?
Dr. Little: The situation probably isn’t that bad, but yes, there is a fair amount of misdiagnosis when it comes to Parkinson’s. We get into some complex questions about what Parkinson’s disease is, actually. There’s a diagnostic standard called the UK Brain Bank Standard which is commonly used. For example if somebody has tremor on one side of their body and at the same when they’re prescribed L-dopa their symptoms improve, this is usually taken as very strong evidence of Parkinson’s. So we really hope that this technology can help provide additional objective information that can be useful in diagnosis, i.e. diagnostic ‘decision support’. That’s useful because in a lot of cases neurologists don’t have much objective information to go on.
You have eighteen thousand so far right?
Dr. Little: Yes that’s right – the Parkinson’s Voice Initiative has been able to recruit eighteen thousand participants – we now have eighteen thousand different voice recordings. Our estimate is about 12% of those have Parkinson’s.
What do you do with that data?
Dr. Little: In these calls we get anonymous information about the patient. We want to know their age, and we want to know whether they’ve been clinically diagnosed with Parkinson’s because this is our reference. We use that definitive clinical diagnosis in order to develop algorithms to categorize people as healthy or not. With the information supplied on the call, and using voice recordings, we then apply algorithms to the recordings that are looking for specific features in the voice. Different clusters of features we’ve determined in lab studies are useful indicators for making that discrimination. The next stage then is to pass those features to a so called ‘classifier’ – this makes a determination about whether or not this person is at high risk of having Parkinson’s.
So can you use this stand alone or do you see it as supplement only?
Dr. Little: I think it is certainly possible to use this technology outside of a medical context but of course the issue is that if you have a diagnosis or if this technology tells you that you may have a diagnosis then what do you do with that information? Of course it would be unethical to be able to have access to that information but without being able to do something about it. So we see this as being supplementary to existing approaches. I think that there are lots of applications for this. For example there are large areas of the world where the ratio of neurologists to patients is so low that basically most people who have Parkinson’s disease in those regions will never see a specialist. So, being able to extend the reach of very, very overworked clinical staff is certainly one of the things that we hope this technology can address.
What’s your ultimate goal for this study?
Dr. Little: The end goal of this study is to be able to produce a practical technology that can be uses in a medical context – to improve our knowledge about Parkinson’s disease and to improve individual lives by making this kind of objective information readily accessible.
How much longer will this study lasts? Can people call in?
Dr. Little: We would really like people to call in: and anyone can call to contribute their voice recording. It takes about three minutes so it’s a very painless process. There are local numbers in a large number of the countries; you just have to pick the number in your country. We’re going to keep running this, we think, probably for a few more months, but we don’t have a specific cutoff point at this time.
Did you expect this kind of response?
Dr. Little: No certainly not. It’s been an overwhelming response, and it’s been fantastic actually. It really shows that when you allow access to this kind of research and make it very, very easy for people to contribute then people do want to contribute and they will. It is about giving them an opportunity to make a contribution to research.
It would take years and years to get this many participants if you had a clinical trial where people had to come in.
Dr. Little: Yes that’s right, absolutely. In a majority of trials, people have to come in and get some kind of treatment or be assessed physically, and it takes a very long time to reach that level of recruitment.
How long did it take to get eighteen thousand?
Dr. Little: We started in late June; it is now mid October so only a few months.
Dr. Little: Yes, it’s overwhelming as I said!
END OF INTERVIEW
This information is intended for additional research purposes only. It is not to be used as a prescription or advice from Ivanhoe Broadcast News, Inc. or any medical professional interviewed. Ivanhoe Broadcast News, Inc. assumes no responsibility for the depth or accuracy of physician statements. Procedures or medicines apply to different people and medical factors; always consult your physician on medical matters.
Sign up for a free weekly e-mail on Medical Breakthroughs called
First to Know by clicking here.
If you would like more information, please contact:
Max Little, Post Doctoral Research Fellow
MIT Media Lab
To read the full report, Dial-A-Diagnosis? The Parkinson's Voice Initiative, click here.