Boston University, Evidation publish study on voice as a digital marker in detection of Alzheimer’s Disease
Researchers at Boston University School of Medicine (BUSM) and Evidation today announced the publication of study results examining the use of voice-based digital markers and transcription as a potential diagnostic tool for Alzheimer’s Disease.
The study, led by Rhoda Au, PhD, professor of anatomy & neurobiology at BUSM, and published in the Journal of Prevention of Alzheimer’s Disease, found use of voice-based digital markers derived from automated processing methods, combined with clinical information that is commonly collected in a typical patient visit, are promising, scalable tools that may enable early detection of dementia and Alzheimer’s Disease.
“Our goal is to find truly scalable ways to detect dementia related symptoms early and intervene quickly to improve quality of life for the growing patient population affected by neurodegenerative disease,” said Dr. Au. “This research shows relatively low-quality audio recordings may be useful tools in characterizing and diagnosing cognitive impairment much more easily than before.”
The observational, retrospective study included 146 participants from the Framingham Heart Study with consensus-confirmed normal cognition, mild cognitive impairment (MCI), or dementia diagnoses. Researchers trained a logistic regression classifier to predict cognitive status, based on demographic, acoustic, linguistic, and paralinguistic variables, which were tested against actual diagnoses. Findings show that a model incorporating paralinguistic variables (statistics pulled from speech-to-text technology) and acoustic variables with demographic information was a sensitive way to detect MCI and dementia, comparable in performance to the use of linguistic variables, which require higher quality audio and manual transcription
“This study shows how paralinguistic variables obtained from off-the shelf hardware combined with automatic speech recognition technology may achieve comparable performance to full-blown language models that require labor-intensive manual transcription in the identification of dementia. This affirms the potential to leverage voice as a non-invasive, inexpensive, and effective way to better characterize and detect neurodegenerative diseases including Alzheimer’s Disease,” said Luca Foschini, PhD, a study author and co-founder and Chief Data Scientist at Evidation.
Alzheimer’s Disease is the leading cause of dementia in older adults and its prevalence is increasing. Diagnosis of Alzheimer’s and other heterogenous neurodegenerative diseases is notoriously complex, disparate, and often completed late in disease progression. Digital markers like voice are promising tools for dramatically reducing current time to diagnosis and may comprise useful screening tools in the future.