February 3, 2023 – The artificial intelligence that may write essays and pass tests may additionally help detect dementia.
Researchers at Drexel University in Philadelphia used the AI behind ChatGPT (which made headlines for writing credible term papers and passing bar exams) to investigate speech, and the system appropriately identified Alzheimer's patients in 80% of cases, in line with the study published within the journal PLOS Digital Health.
The researchers used GPT-3, the language model that powers ChatGPT, to investigate audio clips of individuals describing an image in a regular test for dementia.
Alzheimer's patients often repeated themselves, deviated from describing the content of the image, didn't finish thoughts, and referred to things vaguely as “thing” or “something.”
“GPT-3 is able to detect such a subtle difference that is reflected in the text,” says study writer Hualou Liang, PhD, professor of biomedical engineering at Drexel University.
The software analyzed (also via software) transcribed text from 10-second recordings of healthy adults and Alzheimer's patients. The text trained the GPT-3 model to acknowledge the subtle differences between normal speech and the speech of an individual with cognitive decline.
The GPT-3 machine learning models understand passages of text by converting words into mathematical representations called “embeddings”. The embeddings are multidimensional signals that allow the AI to detect subtle differences and similarities that even experienced doctors cannot hear. GPT-3 compares the passages of text by measuring the space between these signals within the embeddings.
Because GPT-3 only analyzes written text, the method bypasses pauses and other non-word sounds in spoken language. In this case, that turned out to be a bonus: GPT-3's evaluation outperformed some machine learning models developed by other labs that took these sounds under consideration.
However, other studies have found that the “ums” and “ahems” in language could also be vital in detecting Alzheimer’s. Study 2021 that encoded these pauses enabled a machine learning model to detect Alzheimer’s with 90% accuracy, and a separate study A study was conducted in Slovenia that combined text and acoustic features and achieved an accuracy of 94%.
“The best combination is usually to combine both types of features,” says Dr. Frank Rudzicz, associate professor of computer science on the University of Toronto. “There is a lot of information in the words and structure of the transcripts, but also in our tone of voice.”
Detecting Alzheimer's using the voice
More and more researchers are investigating Voice as a biomarkera technique for detecting various diseases, including Alzheimer's.
Worldwide, Alzheimer's cases are only successfully diagnosed in 48% of cases. According to estimates by the World Health Organization. In higher-income countries, the diagnosis rate is 54%, while in low- and middle-income countries only 24% of Alzheimer's cases are detected.
Researchers in the sphere hope to fill that gap by developing a tool that may detect Alzheimer's early – when the consequences could also be too subtle for a physician to detect. “There is no cure for Alzheimer's disease yet, but there are life changes that can delay some of its effects, so early diagnosis is still important,” says Rudzicz, co-founder of a mobile speech evaluation app called Winter light“These types of technologies could also be used for other diseases, including Parkinson’s, depression and so on.”
Doctors could use a tool or computer program in some unspecified time in the future of their practice to check a patient's cognitive abilities. Brain scans or other clinical tests could then confirm the Alzheimer's diagnosis.
Another application could use smart devices like Alexa and Siri to watch your normal conversations (together with your consent) and warn you when it notices troubling word errors. It could even detect other mental health issues like depression and stress.
“The analysis could be performed in a privacy-preserving manner once the system is fully operational,” says Liang. “It could therefore have an immediate and significant impact on curbing the dementia problem in the elderly population.”