Melbourne scientists have developed a screening test for Parkinson’s disease using patients’ voice recordings and an AI algorithm.
The software compares the patient’s voice against existing samples of voices of people who have Parkinson’s and those who do not.
The algorithm, designed as a smart phone app, can reveal in just 10 seconds if you might have Parkinson’s disease and should be referred to a neurologist.
The RMIT researchers are seeking ethics approval for a clinical trial, with a view to the app being adopted in a national screening program – akin to the routine age-determined screening for bowel and breast cancer, and heart disease.
Why the voice?
For more than a decade, researchers have been investigating the use of voice biomarkers – pitch, tone, rhythm – as indicators of disease that can be identified by algorithms developed via machine learning.
As we reported in March, Mayo Clinic scientists announced an algorithm that accurately predicted the likelihood of developing serious heart problems based solely on the voice recordings of patients.
In a small study of 108 people, they found that people “with a high voice biomarker score” were 2.6 times more likely to suffer a build-up of plaque in the heart’s arteries.
They were also found to be three times more likely “to show evidence of plaque build-up in medical tests compared with those who had a low score”.
This is a fast-moving area of research, with immense potential in clinical diagnosis because voice biomarkers aren’t simply proving to be accurate, they’re a much cheaper and faster tool than scans such as MRI.
According to a 2021 article in The Lancet – jauntily titled ‘Do I Sound Sick?’ – vocal biomarkers are being developed to diagnose “a multitude of disorders including Parkinson’s disease, cardiovascular disease and post-traumatic stress disorder”.
In May, German scientists published an overview of the field titled ‘Is speech the new blood?’
In other words, will a voice test eventually replace a blood test to screen for a variety of diseases?
Parkinson’s is a good candidate
Parkinson’s disease is a particularly good candidate for voice screening. This is because people with Parkinson’s undergo changes in their voice which are brought about by three symptoms – rigidity, tremor and slowness (known as bradykinesia).
Neurologists develop an ear for these changes, but accuracy is compromised “due to the large natural differences in people’s voices”.
Even robots struggled to deal with different accents and vocal tones.
The new app, from RMIT, may have solved this problem, thanks to a yoga teacher’s suggestion.
Lead researcher Professor Dinesh Kumar, from RMIT’s School of Engineering, told The New Daily that by adopting the A, O and M sounds used in the Hindu meditation chant, the vocal differences were largely ironed out.
In this video, below, Dr Quoc Cuong Ngo demonstrates the chant-like sounds used in screening.
“These sounds result in a more accurate detection of the disease,” Professor Kumar said.
And while Professor Kumar and his team continue to refine the process of levelling the confounding differences of accents and tone, he doesn’t believe that this accuracy rate will be much improved.
He said the app’s accuracy rate sits between 80 and 90 per cent for detecting Parkinson’s disease (PD).
“This isn’t a diagnostic tool. It’s a screening tool,” he said.
For now, anyway. The RMIT app appears to perform as well, and perhaps slightly better, than traditional diagnostic protocols.
A 2016 review found that eight out of 10 patients received a valid diagnosis, and two out of 10 people were misdiagnosed with a clinical diagnosis.
The review concluded: “The overall validity of clinical diagnosis of PD is not satisfying. The accuracy did not significantly improve in the last 25 years, particularly in the early stages of disease …”
There is no blood test. But maybe a voice test will prove to be the way to go.
Severe COVID-19 and pulmonary diseases
The RMIT team has used the same technology to predict whether or not a case of COVID-19 or other pulmonary diseases will become serious.
Professor Kumar said that in patients with pulmonary disease symptoms from COVID-19, “there is also a change in the voice due to lung infection”.
But, as in Parkinson’s screening, “due to large differences in people’s voices, pulmonary disease is difficult to recognise in its early stages”.
Once again, the sounds of the Hindu chant appear to solve this limitation.
Co-researcher Dr Ngo, from RMIT’s School of Engineering, said the technology was faster and better than any similar AI-based approach.
“Our screening test app can measure, with great precision, how the voice of someone with Parkinson’s disease or person at high risk of hospitalisation from COVID-19 is different from healthy people,” he said.
The researchers have published three papers detailing their work. See here, here and here.
