
Parkinsonās disease lacks a conclusive test, so it is generally diagnosed by assessing symptoms. But now, scientists have shown that AI models can identify signs of the condition in a personās voice with more than 90 per cent accuracy, and possibly before the onset of any movement-related issues.
Parkinsonās is characterised by the proliferation of a misfolded form of a protein called alpha-synuclein. It has been suggested that tests could look for clumps of this protein in peopleās spinal fluid or in .
Looking for a low-cost, non-invasive alternative, at the University of North Texas and his colleagues collected 195 voice recordings from 31 people, 23 of whom had been diagnosed with Parkinsonās. Some of these recordings were then used to train four AI models to detect the condition, based on vocal features such as hoarseness and an irregular pitch.
Advertisement
Up to 90 per cent of people with Parkinsonās disease develop dysarthria, defined as difficulty speaking because the muscles used for speech are weak, says at Binghamton University in New York state, who wasnāt involved in the study.
After the models were trained, different voice recordings from the participants were used to validate them. When put to the test on the remaining recordings, the models demonstrated more than 90 per cent accuracy at identifying people with Parkinsonās disease.
Dysarthria seems to be caused by , and in Parkinsonās progression, says Bishop, which makes such vocal changes āan intriguing early marker for Parkinsonās diseaseā.
The voice-based approach āshows real promise as an early screening toolā, says at Rune Labs, a software and data analytics company for precision neurology in California. āBecause it only requires a microphone and an internet connection, this method could be used remotely, at scale, even in places without easy access to neurologists.ā
But Arnold cautions that the dataset used in this study was small. Larger ones, made up of voice recordings with a range of accents, are needed before it can be used as a standard diagnostic, he says.
medRxiv