26 November 2024: Algorithms that can detect subtle voice changes in PD are emerging as a potential new diagnostic tool for Parkinson’s disease, according to researchers from Iraq and Australia.
Speech impairments are often the first indicators of Parkinson’s disease, but traditional diagnostic methods are often complex and slow, delaying early detection.
Researchers from Middle Technical University (MTU) in Baghdad and the University of South Australia (UniSA) have recently published a conference paper reviewing the advancements in artificial intelligence techniques to detect Parkinson’s disease (PD).
MTU Associate Professor Ali Al-Naji, a medical instrumentation engineer and UniSA adjunct, says all the evidence shows that AI-powered voice analysis could revolutionise early PD diagnosis and remote monitoring of the neurodegenerative disorder.
“Vocal changes are early indicators of Parkinson’s disease, including small variations in pitch, articulation and rhythm, due to diminished control over vocal muscles,” Associate Prof Al-Naji says.
“By analysing these acoustic features, AI models can detect subtle, disease-related vocal patterns long before visible symptoms appear.”
AI techniques primarily use machine learning and deep learning algorithms trained on extensive data sets from simple voice recordings from Parkinson’s patients and healthy controls.
These algorithms extract relevant features, such as pitch, speech distortions and changes in vowels, and then categorise the voice recordings with remarkable accuracy – as high as 99% in one research study.
Researchers say that while Parkinson’s has no cure, early diagnosis and intervention can improve quality of life and slow the progression of symptoms.
“As well as detecting Parkinson’s early, AI could also help monitor patients from a distance, reducing the need for in-person visits,” Associate Prof Al-Naji says.
However, researchers acknowledge the need for further studies on larger, more diverse populations.
“Parkinson’s disease detection from voice using artificial intelligence techniques: A review” was presented to the Fifth Scientific Conference for Electrical Engineering Techniques Research in Baghdad. DOI: https://doi.org/10.1063/5.0236188
An accompanying paper – “Voice-based gender classification: A comparative study based on machine learning algorithms” – was presented by Associate Prof Al-Naji and colleagues at the same conference.
