Overview

Researchers at the University of Rochester have created an AI-driven, speech-based screening tool designed to help individuals assess potential signs of Parkinson’s disease, which is recognized as the fastest-growing neurological disability globally.

Key Features of the Tool

• The tool utilizes a web-based screening test that prompts users to recite two pangrams—sentences that include all 26 letters of the alphabet.
• Within seconds, the AI analyzes voice recordings for subtle patterns associated with Parkinson’s, achieving an accuracy rate of nearly 86%.
• While not a replacement for clinical diagnosis, this tool serves as a quick and accessible method to identify individuals who may need further evaluation, particularly in remote areas.

Research Insights

Parkinson’s disease is typically diagnosed by specialists using a combination of family history, neurological examinations, and brain imaging. The study’s authors emphasize the importance of this AI tool in providing a low-barrier option for screening.

Ehsan Hoque, a professor in the Department of Computer Science at Rochester, stated, “There are vast areas in the US and worldwide where access to specialized neurological care is limited. With user consent, commonly used speech interfaces like Amazon Alexa or Google Home could assist in identifying individuals who may require further care.”

Data Collection and Methodology

To develop and validate the tool, researchers gathered data from over 1,300 participants, both with and without Parkinson’s, in various settings, including home environments and clinical visits.

Participants read two sentences aloud, which the AI analyzed using advanced semi-supervised speech models trained on millions of audio recordings. The tool identifies vocal cues that may indicate the presence of Parkinson’s.

Performance and Future Directions

The tool demonstrated an accuracy of 85.7% in tests, indicating its potential as a preliminary screening method. However, it is essential to note that Parkinson’s is a complex disease, and symptoms can manifest through various means, including speech, motor tasks, and facial expressions.

According to Tariq Adnan, a PhD student involved in the study, “Research indicates that nearly 89% of individuals with Parkinson’s exhibit vocal deformities that can signal the disease, making speech a valuable starting point for digital screening.”

Interactive Demo and Further Research

An interactive demonstration of the lab’s screening tests, including the speech analysis tool, is available online. The study was supported by various organizations, including the National Institute of Neurological Disorders and Stroke and Google.