番茄社区app

A machine-learning model identified mild cognitive impairment and dementia from digital voice recordings of neuropsychological tests, an early study showed.

Among 1,084 people in the Framingham Heart Study whose tests were recorded, the average area under the curve (AUC) reached 92.6% for differentiating normal cognition from dementia, 88.0% for discerning normal cognition or mild cognitive impairment from dementia, and 74.4% for distinguishing normal cognition from mild cognitive impairment.

The model used voice recognition to transcribe recordings to text and leveraged natural language processing methods for analysis, reported Ioannis Paschalidis, PhD, of Boston University, and co-authors in .

"We view our work as contributing to the emergence of digital biomarkers for Alzheimer's and dementia," Paschalidis told 番茄社区app. "We believe they're essential to remote, global, automated assessment at scale, helping to identify individuals at earlier stages of the disease who could become prime targets for the next generation of clinical trials."

Last year, a small study in Japan showed that a predictive model correctly identified people with Alzheimer's dementia with 90% accuracy using audio files of phone conversations of 24 people with confirmed Alzheimer's and 99 healthy controls. That research relied on the fact that people with Alzheimer's disease were likely to speak slowly with longer pauses, spending more time finding correct words.

In the current study, however, how people spoke and whether they faltered were less important than the content of what they were saying. "It surprised us that speech flow or other audio features are not that critical; you can automatically transcribe interviews reasonably well and rely on text analysis through AI to assess cognitive impairment," Paschalidis said.

Paschalidis and co-authors encoded test transcriptions into quantitative data, then trained and tested several models. The final model was trained to assess the likelihood and severity of an individual's cognitive impairment using demographic data, the text encodings, and real diagnoses from neurologists and neuropsychologists.

The study also suggested which parts of neuropsychological tests may be important for discerning impaired cognition using AI. The -- a task in which people are asked to label a picture using one word -- led to an accurate dementia diagnosis more often than others.

Improving the model's discrimination between normal cognition and mild cognitive impairment still needs work, Paschalidis acknowledged.

"That will always be harder to discriminate," he said. "We hope that more data and a larger observation period will help increase discrimination ability."

"If data capture how one operates in daily life and are compared with prior data from the same individual, then even mild changes in cognitive function could be potentially identified," he added.

The findings suggest models like this may help diagnose cognitive impairment using audio recordings, including those from virtual or telehealth appointments, Paschalidis and co-authors pointed out. The method can be adapted to other languages and eventually developed into a web-based tool that could offer widespread, cost-effective dementia screening, they said.

The approach needs to be confirmed, the researchers noted. "We are looking to expand the type of digital data we can analyze with AI to get a more complete picture of cognitive function and its potential decline, and add diversity to the tools being used for assessment," Paschalidis said.

Comment