Using AI To Quickly Diagnose Alzheimer’s Disease and Dementia From Voice Recordings

Using AI To Quickly Diagnose Alzheimer’s Disease and Dementia From Voice Recordings

Scientists develop an artificial intelligence program that detects cognitive impairment accurately and efficiently from voice recordings.

The process of diagnosing Alzheimer's disease takes a long time and is expensive. Clinicians must meticulously transcribe, examine, and analyze every response following extensive in-person neuropsychological tests. Boston University (BU) academics have created a new tool that might automate the procedure and eventually enable it to go online. Without requiring a physical appointment, their computational model powered by machine learning may identify cognitive impairment from audio recordings of neuropsychological testing. Recent issues of Alzheimer's & Dementia: The Journal of the Alzheimer's Association included their findings.

Distinguished Professor of Engineering at the BU College of Engineering and coauthor of the research Ioannis Paschalidis adds, "This technique puts us one step closer to early intervention." According to Paschalidis, quicker and more accurate diagnosis of Alzheimer's disease could lead to larger clinical trials that concentrate on people with the disease in its early stages and possibly permit pharmacological therapies that delay cognitive deterioration. It might serve as the foundation for an internet tool that would be accessible to everyone and would boost the number of early screenings.

The audio recordings of more than 1,000 neuropsychological interviews conducted with participants in the Framingham Heart Study were used by the researchers to train their AI model. This ongoing initiative, directed by BU, looks into physiological issues such cardiovascular illness. Using automated internet speech recognition tools—think "Hey, Google!""—and a machine learning method known as natural language processing, which aids computers in comprehending text, their program transcribed the interviews and then encoded them into numbers. Using a combination of demographic information, text encodings, and actual diagnoses from neurologists and neuropsychologists, a final model was trained to assess the likelihood and degree of a person's cognitive impairment.

Paschalidis claims that the algorithm can detect variations between people with mild cognitive impairment and dementia in addition to accurately differentiating between healthy people and those who have dementia. Surprisingly, it turned out that the content of what people were saying was more relevant than the sound quality of the recordings and the way they spoke—whether their speech was naturally flowing or frequently faltered.

Paschalidis, who is also the new director of BU's Rafik B. Hariri Institute for Computing and Computational Science & Engineering, says, "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." The results indicate that their technology could assist physicians in identifying cognitive impairment using audio recordings, including those from virtual or telehealth visits, but the research team still has to test its findings against other sources of data.

Screening before Symptom Onset

The model also sheds light on which aspects of the neuropsychological test may be more crucial than others in identifying whether a person has cognitive impairment. The clinical tests that were conducted are used to divide the exam transcripts into distinct portions in the research team's model. For instance, they found that the Boston Naming Test, in which patients are asked to label an image with one word, is most helpful in correctly diagnosing dementia. As a result, Paschalidis speculates, "clinicians may be able to arrange resources in a way that enables them to undertake more screening, even before the onset of symptoms."

In order to develop an appropriate plan for care and support for patients and their caregivers, early detection of dementia is crucial. It is also essential for researchers developing treatments to stop or reduce the progression of Alzheimer's disease. According to Paschalidis, "Our models can assist doctors in evaluating patients in terms of their potential of cognitive decline and then appropriately tailoring resources to them by undertaking additional testing on individuals who have a higher likelihood of dementia."

Want to Join the Research Effort?

The study team is asking for participants to complete an online questionnaire and an anonymous cognition test; the data will be used to create customized cognitive evaluations and will also aid the team in improving their AI model.

Reference: Samad Amini, Boran Hao, Lifu Zhang, Mengting Song, Aman Gupta, Cody Karjadi, Vijaya B. Kolachalama, Rhoda Au, and Ioannis Ch. Paschalidis, "Automated detection of mild cognitive impairment and dementia from voice recordings: A natural language processing approach," Alzheimers Disease & Dementia, 7 July 2022.

DOI: 10.1002/alz.12721

Samad Amini (ENG'24), Boran Hao (ENG'19,'24), and Lifu Zhang (CAS'22, ENG'22) also contributed to this study, as did Mengting Song, an ENG researcher; Aman Gupta (ENG'21), a research assistant with the BU Center for Information & Systems Engineering; Cody Karjadi (CAS'17, MET'20) of the Framingham Heart Study; Vijaya B. Kol The National Science Foundation, Office of Naval Research, Department of Energy, National Institutes of Health, National Heart, Lung, and Blood Institute contract for the Framingham Heart Study, National Institute on Aging, Alzheimer's Association, Pfizer, Karen Toffler Charitable Trust, American Heart Association, and Boston University all provided funding for the research.

Funding sources include the National Science Foundation, the Office of Naval Research, the National Institutes of Health, the Framingham Heart Study, the Alzheimer's Association, Pfizer, and the American Heart Association.