New computational model can detect cognitive impairment from voice recordings of neuropsychological tests

Diagnosis of Alzheimer’s disease takes a lot of time and money. After performing a lengthy face-to-face neuropsychological examination, the clinician should copy, review, and analyze all reactions in detail. However, researchers at Boston University have developed new tools that can automate the process and eventually go online. Computational models that utilize their machine learning can detect cognitive impairment from voice recordings of neuropsychological tests-; no direct booking required.Their findings are Alzheimer’s Disease and Dementia: Journal of the Alzheimer’s Disease Association..

“This approach brings us one step closer to early intervention,” said Ioannis Paschalidis, co-author of the paper and a prominent professor of engineering at the BU Faculty of Engineering. He found that faster and earlier detection of Alzheimer’s disease could lead to larger clinical trials focused on individuals in the early stages of the disease, enabling clinical interventions to delay cognitive decline. It states. The number of people who will be screened early. “

The research team uses audio recordings of neuropsychological interviews from more than 1,000 individuals to model a model in the Framingham Heart Study, a long-term BU-led project investigating cardiovascular disease and other physiological conditions. I trained. Using automated online speech recognition tools-; think, “Hey, Google!”-; And a machine learning technique called natural language processing that helps computers understand text, transcribing interviews to the program, I encoded it into a numerical value. The final model was trained to assess the likelihood and severity of an individual’s cognitive impairment using demographic data, text encoding, and actual diagnosis by neuropsychologists and neuropsychologists.

According to Paschalidis, the model not only accurately distinguishes between healthy people and people with dementia, but also detects the difference between people with mild cognitive impairment and people with dementia. And it turns out that the quality of the recordings and the way people speak-; whether their speeches are breezy or consistently waning-; less important than what they are saying. did.

“I was surprised that speech flow and other audio features weren’t that important. We could properly and automatically copy interviews and evaluate cognitive impairment based on text analysis by AI,” said BU’s new director, Rafik B. But Paschalidis said. Hariri Institute for Computational Science and Engineering. The team still needs to validate the results against other data sources, but the findings show that their tools use voice recordings, including those from virtual or telemedicine appointments, to diagnose cognitive impairment. It suggests that we can support clinicians in some cases.

Screening before the onset of symptoms

This model also provides insight into which parts of the neuropsychological test are more important than others in determining whether an individual has cognitive impairment. The researcher’s model divides the test transcript into various sections based on the clinical tests performed. They found, for example, the Boston Naming Test, a test in which a clinician asks an individual to label a photo using a single word, is most useful for accurate dementia diagnosis. “This may allow clinicians to allocate resources for more screening, even before symptoms appear,” Paschalidis said.

Early diagnosis of dementia not only enables patients and their caregivers to effectively plan treatment and support, but also for researchers working on treatments that slow and prevent the progression of Alzheimer’s disease. It’s very important. “Our model helps clinicians assess patients for their potential for cognitive decline, and tailoring them by conducting further tests on patients with a high likelihood of dementia. It’s best to coordinate resources, “says Paschalidis.

Would you like to participate in research activities?

The research team is looking for volunteers to conduct online surveys and submit anonymous cognitive tests. The results are used to provide a personalized cognitive assessment and help teams improve their AI models.