Speech algorithms to detect Parkinson's disease

A British mathematician hopes he can speed up the diagnosis of Parkinson's disease with a cheap test that uses speech signal processing algorithms he developed at Oxford University in the UK. Parkinson's Disease is a progressive, devastating neurological disorder that is difficult and slow to diagnose: there are currently no lab tests or biomarkers that can definitively diagnose the condition, which affects more than 6 million people worldwide. The British mathematician is Max Little, who is currently at the Massachusetts Institute of Technology (MIT), Boston, US, where he is a Wellcome Trust-MIT Postdoctoral Research Fellow. Little is talking about how his algorithms can help detect symptoms of Parkinson's at the opening of the TEDGlobal conference, which is running in Edinburgh, Scotland, this week. TED (Technology, Entertainment and Design) is a nonprofit organization that supports and encourages innovators under the age of 40. Algorithms Little, who began his career writing software, signal processing algorithms and music for video games, discovered while working towards his PhD at Oxford, that voice is affected as much by Parkinson's as limb movement, and the symptoms of the disease can be detected by analyzing speech signals using computer algorithms. In an interview reported by the BBC, Little describes how the algorithms work: "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." He said there are a number of reasons that cause voice patterns to change, even smoking or just having a cold can change them, as well as throat surgery. But he thinks the algorithms will be able to spot the difference between these causes and Parkinson's Disease. He said it is more sophisticated than trying to spot a particular tremor in the voice. The algorithms also take into account other measures that put the tremor in the right context, including if the patient has a cold or if other symptoms are present. From the speech pattern, the algorithms calculate a simple "dysphonia" measure of Parkinson's symptom severity on a standard clinical scale used by doctors (the UPDRS, or Unified Parkinson's Disease Rating Scale). Together with his student Athanasios Tsanas at Oxford, Little showed that it was possible to predict, from non-invasive speech recordings, symptoms of Parkinson's on the UPDRS scale "with a few percent error". In a paper published earlier this year in IEEE Transactions on Biomedical Engineering, Little, Tsanas and colleagues describe a study where they tested the accuracy of some of the new algorithms in discriminating Parkinson's Disease (PD) patients from healthy controls. In total, they computed 132 "dysphonia measures" from sustained vowel sounds, and using a database of 263 samples from 43 people,they showed that four subsets of the algorithms "outperform state-of-the-art results", reaching nearly "99% overall classification accuracy using only ten dysphonia features". "We find that some of the recently proposed dysphonia measures complement existing algorithms in maximizing the ability of the classifiers to discriminate healthy controls from PD subjects. We see these results as an important step toward noninvasive diagnostic decision support in PD," they write. Little and colleagues now want to open the scope of their investigation to include more voice samples. Huge Database of Voices Little, who is a TED Fellow, is using his TEDGlobal platform to invite volunteers to phone in and contribute a 3-minute recording of their voice, so that he and his colleagues can build a huge database of 10,000 voice samples to test and refine the algorithms. The database is part of the Parkinson's Voice Initiative (PVI), which could lead to significant improvements in the care of Parkinson's Disease patients, such as drastically reducing clinical visits for checkups, improving individual treatment decisions, and helping to speed up and reduce the costs of recruiting large numbers of volunteers on trials for new treatments. Little and colleagues also hope the initiative will lead to population-scale screening programs that will help search for early biomarkers that spot the signs of Parkinson's before irreparable damage is done. The project is looking for voices from people, including those who don't have Parkinson's, who are willing to contribute a few minutes of their time, anonymously, over the phone. The researchers have set up phone numbers in 10 countries.