- Researchers built an algorithm to predict Parkinson’s disease from short speech samples.
- Their model was able to predict 80-90% of Parkinson’s disease cases.
- They are now developing their algorithm into an app to help identify those at risk of the condition.
Parkinson’s disease (PD) is the
Currently, PD diagnoses happen late in the neurodegenerative process. Earlier identification of symptoms could allow for earlier intervention and thus longer uncompromised working capacity and prolonged quality of life.
Speech acoustic analysis in PD has received growing scientific interest in recent years as a potential diagnostic biomarker. Studies have
Further understanding how PD affects the voice could lead to the development of screening methods that can distinguish between the voices of PD patients and healthy individuals.
Recently, researchers developed an automated screening method that can distinguish between the voices of PD patients and healthy individuals.
The researchers found their model could predict 80-90% of voices from those with PD.
“The goal of making a Parkinson’s diagnosis easy is admirable,” Dr. Clifford Segil, a neurologist at Providence Saint John’s Health Center in Santa Monica, California, who was not involved in the study, told Medical News Today.
“There are not enough general neurologists or movement disorder neurologists in the world at this time. I agree with the authors that this study is not meant to be a substitute for a detailed neurological exam from a neurologist,” he added.
The study was published in MDPI.
For the study, the researchers built an AI deep learning model using publicly accessible vocal data from 50 Italian people: 22 without PD and 28 with the condition.
They then obtained voice samples from 104 people from Lithuania saying the phonetically balanced Lithuanian sentence: ‘Turėjo senelė žilą oželį’ meaning ‘The grandmother had a little gray goat’.
Among the patients, 61 were ages 39–84 and had PD for up to 14 years, while 43 did not have the condition.
After refining their model, the researchers tested its predictive ability on both Italian and Lithuanian speech samples.
In the end, they were able to predict 90% of PD cases among Italian voice samples, and 80% of Lithuanian samples.
Importantly, predictions occurred in real-time- within seconds as opposed to hours after input.
To understand more about the link between voice and Parkinson’s disease, MNT spoke with lead author of the study, Rytis Maskeliūnas, professor at the Faculty of Informatics, Kaunas University of Technology.
“According to research, a large percentage of people with Parkinson’s disease have speech and voice disorders, such as soft, monotone, breathy, and hoarse voice, as well as uncertain articulation. This may be difficult to hear or distinguish in the early stages of the disease (sounds normal to our ears), but it is what our approach looks for in the signal.”
Dr. Jordan Taylor, neurology section chief at the University of Michigan Health-West, who was not involved in the study, also told MNT:
“Speech and voice dysfunction are very common in Parkinson’s patients and are likely caused by degeneration of various structures throughout the brain and brainstem. Amongst other issues, this leads to problems with appropriately activating throat musculature.”
Dr. Segil said, however, that it would be ‘challenging’ to agree with the researchers’ position that early PD affects the vocal cords, diaphragms, and lungs.
“PD patients also can have uncontrollable mouth movement calle[d] ‘oral-buccal tremors’ which involve PD patients mouth and cheeks. These movements are more likely to affect advanced PD patients after diagnosis more than proposed dysfunction of a newly diagnosed Parkinson’s patient’s vocal cords, diaphragm, and lungs.”
— Dr. Clifford Segil
When asked about the study’s limitations, Dr. Taylor noted that as the study only utilized two languages to train the algorithm, the findings may not apply to wider populations speaking other languages.
“While speech dysfunction is common, Parkinson’s patients have a heterogeneous collection of symptoms, some of which may present earlier than others. Using only speech dysfunction as a diagnostic measure may end up missing some early Parkinson’s patients,” he said.
Dr. Lisa M. Shulman, professor of neurology at the University of Maryland School of Medicine, also not involved in the study, told MNT that while the tool may be able to detect vocal changes in people with PD, there is no evidence to show that it can detect speech changes indicative of PD in people prior diagnosis.
Dr. Julie Gerberding, CEO of the Foundation for the National Institutes of Health and former Director of the CDC, also told MNT that a collaborative approach to further research is needed for this research to reach the clinic.
“We need to develop a comprehensive, open-access speech-sample database to enable the development of better biomarkers to identify Parkinson’s disease. Biomarkers are characteristics of the human body that can be measured, like blood pressure or cholesterol level. This is a complex task, and no single organization, acting alone, can successfully address a complex biomedical challenge like Parkinson’s disease,” Dr. Gerberding said.
The researchers hope to develop their findings into an app to help people understand whether they are at risk of developing PD.
They are currently conducting clinical trials to see if different phone models with different microphones—and audio recording quality— impact their analyses.
“This app may serve as a curiosity tool, an early stimulus to learn more about your condition, particularly if Parkinson’s disease runs in your family or just in case. In my opinion, the purpose of this is not to replace a real doctor but rather to motivate you to go and get a professional diagnosis.”
— Prof. Rytis Maskeliūnas
He noted that they are also developing another version for doctors alongside the Lithuanian University of Health Sciences that produces metrics and aids in the screening process.
“If standardized parameters can be created for studying the voice and speech alternations in PD patients—and those parameters ultimately become clinically validated—it could become a very powerful screening tool for early recognition of a disease that impacts millions of people around the world!” Dr. Elana Clar, movement disorders specialist and neurologist at New Jersey Brain and Spine in Oradell, New Jersey, who was also not involved in the study, told MNT.