Adding a Novel Voice to Parkinson’s Research
Thursday, 21st February 2013

Parkinson’s is difficult to diagnose, and a big reason for this is that, as of yet, there is no objective test for determining if a person has the disease. Often, it can take years and visits to various specialists to definitively determine if someone will develop PD. A big hurdle: No biomarker for the disease currently exists. For now, the process of diagnosis remains an inexact science.

Enter Max Little, a PhD and a mathematician whose career revolves around finding patterns in data. Little is developing technologies to provide objective clinical information about PD symptoms that would be valuable in diagnosis and monitoring symptom progression, using voice and other measurements collected through telephones or smartphones.

Little’s work has gotten a good amount of press of late, thanks to a 2012 TED talk he did on his technology, and a recent column in the Huffington Post. He says that in laboratory-controlled trials, he was able to detect Parkinson’s with 99 percent accuracy, all from analysis of a short voice recording.

His Parkinson’s Voice Initiative seeks to confirm his original findings by testing thousands of voice recordings obtained using telephone calls from around the world. The study will allow Little to learn more about the efficacy of the technology outside of a laboratory setting, with the end goal of one day being able to provide objective clinical information to as many people as possible.

Little isn’t the only scientist to concentrate on recording and analyzing voices as a means to detect and track Parkinson’s: Rahul Shrivistav, PhD, of Michigan State University recently found that his technology interpreting small changes in speech that take place early on in PD is effective in diagnosing the disease more than 90 percent of the time.

The reality is, we’re likely years away from being able to use voice recordings to diagnose Parkinson’s, and certainly, before a smart phone app that can tell people whether they might go on to develop the disease. And even in this potential future, a physician will always be a critical part of the process.

Still, Little’s work is an intriguing example of using creative technologies and patient-derived data to improve how we diagnose and treat people with PD. Here, we sit down with him to discuss his projects, and where he hopes that they might lead.

MJFF: Tell us more about how you believe voice could predict PD.

ML: My major project during my PhD studies at Oxford was to develop algorithms for detecting various voice disorders. I found that these algorithms were particularly successful in detecting Parkinson’s disease. Specifically, we looked into how people’s voices change as quantified through a very large number of measures of voice features such as vocal tremor, the “breathiness” of a voice, and vocal timbre.

I also learned that as of now, there is no objective test for detecting and quantifying a person’s PD symptom severity. I have a good friend with PD, so I wanted to think more about how I could help. So I decided to concentrate more fully on Parkinson’s, and I advanced these initial findings into more extensive lab work analyzing the voices of people with PD. This is when we were able to detect Parkinson’s with up to 99 percent accuracy based on these algorithms. We also found that we could predict an individual’s severity of symptoms based on their voice and on the Unified Parkinson’s Disease Rating Scale (UPDRS), the tool most widely used by clinicians.

MJFF: So most of your analyzed data to date was collected in a controlled setting.

ML: That’s correct. Certainly, in a clinical environment you are better able to monitor the quality of the recordings themselves, and also make easier comparative analyses based on a person’s clinical rating. In fact, we continue to find ways to make our results applicable to outside, ongoing clinical studies. We’re working closely with [MJFF awardee] Connie Marras, MD, PhD, at the University of Toronto, who’s conducting studies around mutations in the LRRK2 gene. By applying our voice-based approach to her study populations, we hope to learn more about who might be more susceptible to PD, and to find ways to predict risk, and then detect the disease.

MJFF: You’re now collecting voice recordings via telephone for your Parkinson’s Voice Initiative. Could this really work over the phone?

ML: I’m optimistic that it could. The good news is, it’s easy to collect data via telephone. The investment into research for our generous volunteers is less than it would be by asking them to come in to a clinic. Of course, there are issues with maintaining uniformity of recording, and with quality control, but we believe we already have an outstanding 12,000 usable samples (out of 17,000 collected) to implement into our study. We have mountains of data, and this is a good thing when trying to solve a complex problem like finding a way to detect PD.

MJFF: How close are we to being able to use voice as a sort of “dry biomarker?”

ML: Well, there are still some critical experiments needed to demonstrate this. In principle, we’ve done enough work to show that we can detect PD through voice at the point where someone would show the first clinical signs of Parkinson’s. But several new experiments would be necessary to be able to home in on using voice as a predictor of Parkinson’s in people who do not yet show symptoms that a movement disorder specialist would identify as PD.

MJFF: What’s next for you in your work?

ML: In addition to this study with the University of Toronto, I’m also working with Ray Dorsey, MD, at Johns Hopkins University to see how well voice recordings and other digital measurements collected by smartphones can replicate UPDRS measures. And our team now has the monumental task of analyzing these 12,000 plus voice samples to prove finally whether it is possible to use the telephone to detect PD, or not.