Symptoms can be detected by computer algorithms that analyse voice recordings.

You may not think applied mathematics has much of an application when it comes to neurological diseases but Max Little has shown how helpful algorithms can be in detecting Parkinson’s. There are currently no biomarkers, such as a blood test, that can identify whether someone has the disease.

This means that the sufferer is often only diagnosed with Parkinson’s when symptoms are full blown. In response, Little has set up the Parkinson’s Voice Intiative which uses the sound of one’s voice, over the phone, to test whether they have the degenerative disease. They are currently building a database of people’s voices and if tests go well they plan to implement this form of testing as a viable and cheap way of diagnosing the disease.

Applying mathematics

Little came up with the idea when he was working on his PhD in applied Mathematics at Oxford University. He says, “In applied maths the goal is to come up with something that is mathematically interesting but at the same time useful in practice. I was searching around for applications I could use and I came across this interesting area of biomedical voice analysis: analysing voices in order to try and extract clinically relevant information. I started looking at voice disorders but it wasn’t until I had a chance encounter with a researcher in 2006 that I realised these techniques might actually be useful in Parkinson’s disease”

Testing for symptoms

The team initially carried out voice recordings in the lab but Little was keen to test whether it works by phone as this will allow people worldwide to test for the disease. It works by applying signal processing algorithms to the phone recording. A machine learning algorithm then makes a decision about whether or not the person has Parkinson’s. According to Little there are several vocal symptoms of Parkinson’s that can be detected in the recordings.

Having this technology which can diagnose the disease and detect it early would be a massive boon for Parkinson’s disease worldwide

He says, “One class of symptoms is to do with tremor. If I ask you to say ‘ahh’ and keep the pitch steady then naturally your pitch will fluctuate a little bit but for those with Parkinson’s the fluctuation tends to be quite large and uncontrollable. The second class of symptoms is to do with the vocal folds. Typically, in a normal healthy voice the vocal folds snap together robustly at the end of every cycle – most of the energy in your speech is generated by this mechanism.

People with Parkinson’s often have difficulty keeping their vocal folds held tightly together and their voice can sound weak and breathy. The third class of symptoms we are looking at are to do with the timbre or tone of your voice. It turns out that in many cases, people with Parkinson’s disease have an inability to keep their jaw, tongue or lips steady when you ask them to do something like sustain a sound of constant timbre for a length of time.”

Detecting the disease early

While Parkinson’s is an incurable disease, there are drugs that can alleviate symptoms and which are more effective if taken earlier before symptoms become severe.

Little says, “The sad thing is that many people who have Parkinson’s probably aren’t being treated, and there are many people who haven’t been diagnosed, so having this technology which can diagnose the disease and detect it early would be a massive boon for Parkinson’s disease worldwide.”

In applied maths the goal is to come up with something that is mathematically interesting but at the same time useful in practice

Little is aiming to get 10,000 voice recordings by phone. If the experiment is successful then this form of cheap, remote diagnosis could greatly improve the lives of those who suffer from the disease. Anybody, with or without the disease can ring up and take the test.


If you want to help out you can get the correct number for your country of residence at www.parkinsonsvoice.org.


Watch Little’s TED talk below:


 

The End
404
There were no articles found matching your filter
Loading...
READ MORE