AI Detects Dementia in Voice Recordings
BY GINA MANTICA
There are over 10 million new cases of dementia each year worldwide, according to the World Health Organization. Researchers conduct neuropsychological tests to assess patients’ cognitive functioning, but transcribing the audio recordings takes a lot of time. Artificial intelligence (AI) algorithms can be used to analyze voice recordings of individuals to detect their dementia status efficiently and accurately.
A team of Boston University researchers led by Vijaya Kolachalama, an Artificial Intelligence Research Initiative Affiliate at the Hariri Institute and Assistant Professor of Medicine at the BU School of Medicine, developed a digital tool that assesses patients’ speech for signs of dementia. The real-time screening tool could allow for earlier detection and treatment of Alzheimer’s disease. The findings were published recently in the special issue focused on AI & Dementia in Alzheimer’s Research and Therapy journal.
Kolachalama and colleagues used over 1,000 voice recordings of neuropsychological tests administered to 656 participants in the Framingham Heart Study. The study has been ongoing since 2005, and the cognitive status of participants is documented over time. This means that there are multiple recordings of some participants taken when they have different levels of cognitive functioning, ranging from normal cognition to severe cognitive impairment. The researchers realized that they could use these voice recordings to train an AI algorithm to detect dementia in audio. “With the emergence of speech recognition and analysis tools, there was a realization that these recordings used for quality control were now data in of themselves because speaking is a complex cognitive skill,” said Kolachalama.
The AI algorithm was able to distinguish between voice recordings of participants with dementia, and those with normal cognition. This digital health tool allows for an automated analysis of a person’s cognitive status, without the need for self-reporting symptoms. And, Kolachalama and colleagues’ AI algorithm does not require manual processing or transcription of voice recordings. “Our study is the first to show that AI algorithms (with minimal data processing) can be developed to analyze voice recordings of individuals to detect dementia status,” said Kolachalama.
There is an urgent need to identify reliable, affordable, and easy-to-use strategies for the detection of signs of dementia. Early detection can help families plan future care and treatment, sometimes delaying or preventing the progression of dementia. “Creating a digital tool that can simply assess voice recordings to detect dementia can have broad implications,” said Kolachalama, “Such tools have the potential to be deployable in various, non-clinical settings.”
Interested in learning more about the research happening at the Hariri Institute? Sign up for our newsletter here.