Will Mild Memory Loss Progress to Alzheimer’s Disease?

Boston University researchers reveal an easy-to-administer AI-driven method that predicts the likelihood of developing Alzheimer’s Disease within six years 

By Maureen Stanton

As we get older, we often dismiss forgetfulness as a normal aspect of aging, and for many it is. However, mild cognitive impairment (MCI), a condition where people have noticeably more memory or thinking problems than what’s expected, is a high-risk factor for Alzheimer’s Disease (AD). MCI can be caused by many health factors but 3-15% of individuals with MCI will progress to AD each year. 

Through a news story published today in The Brink, Boston University researchers announced today that they have developed an easy-to-use AI-driven tool that identifies those with MCI who are likely to develop AD within six years. Their goal is to open the window to new FDA-approved treatments that can slow the disease’s progression when caught early.  

“If you think about many other chronic diseases including diabetes, hypertension, even cancer, there are early diagnostic procedures and even predictive models used for prognosis,” says Ioannis (Yannis) Paschalidis, lead author, distinguished professor of engineering, and director of the Hariri Institute for Computing at Boston University. “This has not been the case for Alzheimer’s disease. Our study makes an important step in that direction.” 

In contrast to earlier studies that have sought to predict the conversion from MCI to AD using expensive and invasive traditional AD biomarkers such as PET and MRI scans, Boston University scientists have developed an AI-driven method that relies only on speech-to-text modeling and basic demographic data to achieve highly predictive results. This approach provides an opportunity to develop an inexpensive and easy-to-use screening tool, facilitating the potential for remote assessment and democratizing access to this type of screening.   

Study results published June 26, 2024 in Alzheimer’s & Dementia: The Journal of the Alzheimer’s Association showed this method achieved an accuracy of 79% and a sensitivity of 81% in predicting MCI-to-AD conversion in six years.

“Speech is a ubiquitous modality and reveals a lot about a person’s cognitive state, from sentence composition to the way they structure their thoughts and dialogue,” says Paschalidis.

The novel method uses Natural Language Processing and sophisticated Large Language Models (LLMs), a technology popularized with the introduction of ChatGPT.  In the study, speech recognition (like the one used by Amazon’s Alexa or Apple’s Siri) was used to  generate human language text from audio recordings of neuro-psychological tests conducted with 166 Framingham Heart Study participants. LLMs then encoded the text into numerical vectors and machine learning combined those vectors with basic demographic information to stratify the risk of developing Alzheimer’s disease in the future. 

“A major strength of this study is its use of semantic features from the structured speech-to-text data,” says Paschalidis. “This has the potential to transfer this entire AI pipeline into many different dialects and languages, leveraging powerful Natural Language Processing models in different languages.”

The study results suggest that older women with lower education levels and those carrying one or two copies of the apolipoprotein E (ApoE) E4 allele, are more likely to progress to AD. This is consistent with earlier research suggesting  that individuals who inherit one copy of ApoE E4 genotype have a high risk of developing AD, while those who inherit two copies have an even higher risk. Women who progressed to AD averaged 1.4 years older than males, suggesting that females may be more prone to progression due to their longer life span. 

NIH reports that about 6.7 million Americans aged 65 and older are living with Alzheimer’s dementia today, and that this number could grow to 13.8 million by 2060 barring the development of medical breakthroughs to prevent, slow, or cure AD.  

“This study, with a focus on predicting the likelihood of developing AD, builds on our earlier work using Natural Language Processing methods to create a screening tool that identifies different stages of dementia based on automated transcription of digital voice recordings,” says Paschalidis.  “With continued development and refinement, our predictive model may contribute to early intervention and selection in clinical trials for novel AD treatments, ultimately improving patient outcomes.” 

This research was funded, in part, by the National Science Foundation, the National Institutes of Health, and the BU Rajen Kilachand Fund for Integrated Life Science and Engineering.

Press Contact: Maureen Stanton, Boston University Hariri Institute for Computing. Email: Stanton@bu.edu. Phone: 617.358.5973