Friday, August 31, 2012

In the News: New Software Can Diagnose Parkinson's Disease Simply By Listening To Your Voice

by David J. Hill
Contributing Writer, Singularity University.

Software like Apple's Siri that responds to your voice is convenient and incredibly cool, but what if a similar kind of voice analysis could diagnose disease? A new initiative aims to make detection of Parkinson's disease as easy as making a phone call. Computer algorithms developed by TED Fellow and applied mathematician Max Little can analyze vocal recordings for characteristic anomalies in an individual's voice brought on by the disease. The noninvasive method can detect Parkinson's with 86 percent accuracy in blind testing of 50 voices, and the rate increases to 99 percent when individual's have mid to late stage Parkinson's.

The Parkinson's Voice Initiative has been formed to create a database of 10,000 voice recordings from people all over the world to improve the algorithms. The goal is to make the technology available to doctors within the next two years.

As Little told the BBC, "This is machine learning. We are collecting a large amount of data when we know if someone has the disease or not and we train the database to learn how to separate out the true symptoms of the disease from other factors." The effort was announced at this year's TEDGlobal.

Here is Max Little presenting the Initiative:

Unfortunately, there are no known biomarkers for Parkinson's, so diagnostic testing focuses on evaluating degrees of tremors in the clinic. Along with these characteristic tremors, voice changes are common, such as whispering, breathiness, and a shift to higher tones. In fact, the voice can be weakened by as much as 10 decibels compared to an average speaker. Though these changes can go unnoticed by a person with a disease, relatives and friends more often pick up on it, but a computer algorithm that detects nuances of speech and subtle changes could detect abberations with much greater frequency.

The technology can scale easily as patients can do the tests themselves in minutes and are as cheap as making a local phone call. This could not only help screen people for early stages of the disease, but it could also allow doctor's to track the disease progression in patients and therapists to monitor the effectiveness of voice therapies without patients having to come into the clinic. This could save valuable resources and allow healthcare workers to have more frequent checkups on patient health remotely.

To get a sense of the change in voice that occurs, check out this video from the Parkinson Voice Project, a nonprofit organization that uses intensive therapy to help individuals with Parkinson's and other neurological disorders regain their voices:

The Initiative hopes to use the software to help with patient treatment, so that drug dosage and timing could be optimized. Additionally, clinical trials could benefit from classification methods more accurate than current methods that may fail to detect some with Parkinson's. With a database of recordings available for analysis, more sophisticated algorithms could also be developed that may lead to a scoring system for disease progression based on voice alone.
The potential of this technology goes beyond Parkinson's disease as voice changes can be caused by other neurological diseases, such stroke, multiple sclerosis, or Lou Gehrig's disease (ALS), as well as cancer that affects the throat (larynx, esophageal, neck, and even lung cancer). Voice changes also occur with viral and bacterial infection, like the common cold or flu, and are one of the characteristics of heavy smoking. So if similar types of voice recording pools could be collected and analyzed with these algorithms, the potential to detect diseases and monitor their progression could be developed.

As Gizmodo suggests, voice-analyzing algorithms might someday be integrated into a smartphone app like Siri. Even a simple app that could monitor your regular voice commands or conversations on your smartphone and pop up an alert if a significant change is detected could be a valuable tool in the coming wave of healthcare apps aimed at preventative care.
(If you are interested in participating in the study, you can find the number to call at the PCI website.)

This material published courtesy of Singularity University.

No comments:

Post a Comment