Can machine learning slow cognitive decline?

Boston University researchers developed a model that detects cognitive impairment accurately and efficiently from voice recordings.

BY GINA MANTICA

Alzheimer’s Disease diagnoses are timely and expensive, requiring hour-long in-person neuropsychological exams with trained clinicians that then transcribe, review, and analyze individuals’ responses. But researchers at Boston University developed a tool that could automate analyses and eventually, enable online data collection to diagnose dementia. Fast and early detection could enable more widespread clinical interventions that slow cognitive decline. It can also enable next-generation, larger clinical trials that focus on individuals in early stages of the disease.

“This approach brings us one step closer to early intervention,” said Yannis Paschalidis, incoming Director of the Hariri Institute and co-author of the recent publication. “It can form the basis of an on-line tool that could reach everyone and could increase the number of people who get screened early.”

Paschalidis, a Distinguished Professor of Engineering, teamed up with faculty members from the School of Medicine, including Hariri Institute Research Fellows Vijaya Kolachalama and Rhoda Au, to develop a machine learning model capable of detecting cognitive impairment from audio recordings of neuropsychological tests. The findings were published recently in Alzheimer’s & Dementia: The Journal of the Alzheimer’s Association.

The team trained their model using audio recordings of neuropsychological interviews from over 1,000 individuals that participated in the Framingham Heart Study. They leveraged automated online tools for voice recognition (e.g., Google voice) and adapted a machine learning technique called natural language processing to analyze the resulting text automatically. Paschalidis and colleagues compared these transcripts to 133 manual transcriptions to ensure their accuracy. After transcription and encoding the text with natural language processing, the model was trained to assess the likelihood and severity of an individual’s cognitive impairment.

The model not only distinguishes between healthy individuals and those with dementia accurately, but also detects differences between those with mild cognitive impairment and dementia. “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,” says Paschalidis, referring to the emerging power of artificial intelligence to replace more mundane tasks now done by specialists and taking time and effort. Though the model still needs to be validated on other datasets, the team’s findings suggest that it can support clinicians in diagnosing cognitive impairment using audio recordings, including those from virtual or “telehealth” appointments.

Additionally, the model provides insight into what parts of the neuropsychological exam might be more important than others in determining whether an individual has impaired cognition. The researchers’ model splits the neuropsychological exam transcripts into different sections based on the clinical tests performed. They discovered that the Boston naming test, during which clinicians ask individuals to describe a picture using one word, is most informative for an accurate dementia diagnosis. “This might enable clinicians to allocate resources in a way that allows them to do more screening, even before symptom onset” says Paschalidis.

Early diagnosis of dementia is not only important for patients and their caregivers to be able to create an effective plan for treatment and support, but it is also crucial for researchers to work with subjects to develop therapies that prevent Alzheimer’s Disease progression. “Our models can help clinicians assess patients in terms of their chances of cognitive decline, and then best tailor resources to them by doing further testing on those that have a higher likelihood of dementia,” says Paschalidis.

The BU team has created a website (health-ai.bu.edu) encouraging volunteers to take a survey and submit a test anonymously, in an effort to provide personalized cognitive assessment and collect more data that can help further research.

To learn more about the Hariri Institute’s transformational research, click here to sign up for our newsletter.