Could AI Help Clinicians to Predict Alzheimer’s Disease Before It Develops?

The future of AI for Alzheimer’s.


Alzheimer’s disease is a devastating disease that begins with vague, often misinterpreted signs of mild memory loss followed by a slow, progressively severe decline in cognitive ability and quality of life.

Currently, there is no effective cure or prevention for this crippling disease, which causes emotional turmoil for both patients and their families.

Due to the nature of Alzheimer’s disease and how it takes hold within the brain, likely, the best way to delay its onset and slow its progression is through early intervention. In other words, the earlier clinicians can detect Alzheimer’s disease —even before symptoms begin to appear—the more likely they can potentially one day be able to delay and be able to treat it effectively.

In partnership with our colleagues from Pfizer, IBM scientists have developed a potential AI model that could help predict Alzheimer’s disease’s eventual onset within healthy people with 71 percent accuracy.

The model uses natural language processing to analyze one to two-minute speech samples from a brief, clinically administered cognitive test. These short samples of language data were provided by the Framingham Heart Study, a long-running study tracking various aspects of health in more than 5,000 people and their families since 1948.

The Framingham Heart Study is a community-based, multi-generational, and longitudinal cohort study initiated in 1948 to study various aspects of participants’ health. Its datasets are some of the most often used and cited in health research, with nearly 4,000 research publications derived from its data as of the end of 2019.

Speaking with TechExplorist, Guillermo Cecchi, principal researcher for computational psychiatry and neuroimaging at IBM, said, “This new combined data-driven method, some of which are already available, with the algorithmic formalization of existing knowledge in psycholinguistics, which in most cases is not.”

“The goal of this work is to provide a tool to monitor progression, including the effects of a potential treatment, in a way that is non-burdensome, simple, and inexpensive.”

Cecchi believes that together with other sensing streams aside from speech (e.g., accelerometers), AI can help identify subtle patterns consistently observed in the affected population or, in particular, sub-cohorts to be tied to existing biological knowledge.

While explaining his work, he said, “A good analogy is an electrocardiogram: this is the ‘heart’s voice’ and is one – but not the only – tool a cardiologist uses to judge your heart health. It is nevertheless essential, and now smartwatches are offering it as a feature that can be tracked continuously. Our work is similar to finding these subtle patterns with consent, only that the dimensionality of the ‘mind’s voice’ is much larger, as it includes text structure, text content, acoustics, and prosody.”

During this study, Cecchi and the team wanted to understand how to constrain their analysis such that it would be clinically relevant and not just another AI study. Combining data-driven methods with the algorithmic formalization of existing knowledge in neurology and psychiatry created this new approach.

For the experiment, scientists applied their linguistic model to single language test samples collected while the participants were considered cognitively healthy. Their model successfully predicted whether they would develop Alzheimer’s before the age of 85 with an area-under-the-curve (AUC) of 74% (while chance is 50%).

“Finding that our models can significantly predict conversion to AD is very exciting,” Cecchi told Techexplorist.

Elif Eyigoz, a research staff member at IBM, said, “Psycholinguistic studies on AD and natural language processing methods have provided the technical background for this study. Most importantly, however, this study was made possible by the collection of this particular dataset. Other studies detect and diagnose AD in already diagnosed patients or in subjects who already developed cognitive impairment due to aging. However, this dataset includes observations from subjects while they were cognitively healthy.”

“It also includes evaluation of these subjects’ cognitive status years later by clinicians. Studies like this one require data collection efforts spanning many years, as diseases like AD develop very slowly in decades.”

Language impairment in Alzheimer’s disease:

Prominent dysfunctions are agraphia, telegraphic speech, and repetitiveness. Agraphia means the presence of significant writing errors, such as misspellings. Telegraphic speech is defined as a form of communication consisting of simple and short sentences, similar to those produced by infants learning to speak and resembling the form used in telegrams to economize length. Repetitive speech involves repetitive questioning, repetitive stories/statements, and repetitive themes. 

Cecchi said, “Additionally, decline in structural complexity of utterances has been observed, along with lack of referential specificity, e.g., saying ‘person’ or ‘woman’ instead of ‘mother’ in the Cookie Theft task used in our study.

Before the symptoms appear, this AI predicts the possibility of Alzheimer’s by extracting all of the features mentioned above and combines partial information from them to determine the likelihood of the onset of AD. 

In the future, Cecchi and his team believe that their work will open new avenues for research in longitudinal monitoring of neurodegenerative diseases using speech as an easy and effective tool, along with other sensing modalities at home and clinical observation. AI will be essential to make sense of such a disparate collection of data.

Eyigoz reported, “Through this study, we want to encourage the collection of large-scale longitudinal datasets of language production beginning from middle-age. This may sound very simple, but you need to understand that if we would start collecting this data today, it would be useful after 40 years, in 2060, for predicting the future onset of AD in middle-aged people.”

“Collection of large-scale datasets spanning decades is not ocostlysive, but also involves serious legal complexities due to privacy and security issues about accessing medical records and also about collecting speech recordings. Still, I believe it can be accomplished.”

Journal Reference:

  1. Elif Eyigoz et al. Linguistic markers predict the onset of Alzheimer’s disease. DOI: 10.1016/j.eclinm.2020.100583