US scientists will develop software with artificial intelligence, Voice as a Biomarker of Health, capable of analysing patients’ voices to diagnose and study diseases.
Experts from 12 research institutes led by the University of South Florida (USF) will assemble a training database of people’s voices while ensuring privacy. Using the dataset, they will train an AI system that doctors can use to identify potential diseases and neurological disorders by analysing a person’s speech.
The software will be trained to detect five types of disorders:
- vocal disorders — malignant and benign lesions of the larynx, and vocal cord paralysis;
- neurological and neurodegenerative disorders — Alzheimer’s disease, Parkinson’s disease, stroke, amyotrophic lateral sclerosis;
- affective and psychiatric disorders — depression, schizophrenia, bipolar disorder;
- respiratory diseases — pneumonia, chronic obstructive pulmonary disease;
- pediatric voice disorders — speech or language development delay, autism.
“These five categories of diseases were chosen based on publications by researchers in the field of voice AI over the past 20 years,” said the project lead and assistant professor of the USF Department of Otolaryngology Yael Bensusan in an interview with The Register.
In her view, voice is one of the simplest and cheapest biomarkers to study. Collecting such data does not pose physical risks to patients and can be carried out in settings with limited resources, especially with the use of modern technologies, added Bensusan.
To safeguard privacy and security, experts will employ federated learning with the backing of Owkin, an AI biotechnology startup.
The U.S. National Institutes of Health (NIH) will commit about $14 million to the project over four years. If approved by Congress, it could receive additional funding.
In the first year, NIH will provide $3.8 million to build a large and diverse database of patients’ voices in clinical settings. Using it, researchers will train the AI to recognise common features in the sound of speech of people diagnosed with certain diseases.
Voice as a Biomarker of Health forms part of a broader NIH initiative to accelerate the adoption of artificial intelligence in healthcare. Over four years the agency has pledged to invest up to $130 million in medical AI projects for data collection, tool development, and resources to train researchers.
In February 2022, Neuralink, Elon Musk’s company, were accused of cruelty to animals during neural-implant testing.
In July, the AlphaFold algorithm from DeepMind’s AI lab predicted the structures of more than 200 million proteins.
In June 2021, the WHO published the first global report on the use of AI in medicine and outlined the main principles for its development.
Subscribe to ForkLog News on Telegram: ForkLog AI — all the news from the world of AI!