AI algorithm detects signs of Alzheimer’s disease through language
With no cure and no straightforward way of diagnosing the disease, scientists are exploring every avenue when it comes to detecting Alzheimer’s during its early stages. One group of researchers has turned its attention to subtle differences in the language of sufferers, and have developed an AI tool they say can pick up on these as a way of potentially screening for the disease.
The research was carried out at New Jersey’s Stevens Institute of Technology and focuses on the way some Alzheimer’s sufferers express themselves. The disease, and others that cause dementia, can impact some parts of the brain that control language, meaning that sufferers can struggle to find the right words, perhaps using the word “book” to describe a newspaper, or replacing nouns with pronouns, for example.
“Language deficits occur in eight to 10 percent of individuals in the early stages of Alzheimer’s disease (AD), and become more severe and numerous during its later stages,” lead author of the study, K.P. Subbalakshmi explains to New Atlas. “Note that this statistic is only valid for the early stages of AD. That said, it is standard practice in clinical settings to use language as a way to screen for AD.”
Subbalakshmi and her students set out to develop an AI tool that could detect these language differences, by turning to a standard picture description task currently used in language screening for Alzheimer’s. This asks subjects to describe a drawing of children stealing cookies from a jar, and the team drew on existing transcripts of more than 1,000 interviews, from both Alzheimer’s patients and healthy controls.
These texts were used to train the AI algorithm, with the individual sentences broken down and assigned numerical values so the system could analyze the structural and thematic relationships between them. Over time, this enabled the algorithm to learn to distinguish between the sentences spoken by healthy subjects and Alzheimer’s sufferers with more than 95-percent accuracy, according to the team. Furthermore, it can also explain why it came to the conclusions that it did.
“This is a real breakthrough,” says Subbalakshmi. “We’re opening an exciting new field of research, and making it far easier to explain to patients why the AI came to the conclusion that it did, while diagnosing patients. This addresses the important question of trustability of AI systems in the medical field.”
From here, the team hopes to expand the tool for use in languages other than English, and even enable it to diagnose Alzheimer’s using other types of text, such as an email or social media post. The researchers also see great potential in using it to track how the disease progresses over time, as a way of detecting it in its very early stages.
“We are exploring other angles of interpretability of the AI and also looking to find datasets that has a time evolution factor to it,” explains Subbalakshmi. “That is, a dataset that tracks patient’s language abilities as time progresses. This will help us develop personalized detection for individuals that are at risk of developing this disease.”
The researchers presented their research at last month’s 19th International Workshop on Data Mining in Bioinformatics.