Predicting Alzheimer鈥檚 earlier
Photos by Ashley Barnas Larrimore and courtesy of Alyssa Lanzi December 09, 2025
91原创鈥檚 DementiaBank outshines global competitors, driving machine-learning advances in early dementia prediction
Could the way we speak 鈥 from subtle stutters to repeated words 鈥 reveal the earliest signs of Alzheimer鈥檚 disease? Researchers and data scientists worldwide think so, and their quest for answers just led to a major win for the 91原创.
Alyssa Lanzi, assistant professor of communication sciences and disorders (CSCD) in 91原创鈥檚 College of Health Sciences, leads efforts to expand and diversify , a shared database of multimedia interactions for studying communication in dementia.
Her research, supported by a $3.7 million National Institute on Aging (NIA) grant, has recently earned national recognition after being selected among thousands of databases for use in the NIA鈥檚 Pioneering Research for Early Prediction of Alzheimer's Disease & Related Dementias EUREKA . The two-year challenge aims to develop novel and inclusive approaches for the early prediction of Alzheimer鈥檚 disease and related dementias through three phases that build upon one another.
Phase I of the challenge, called FindIT!, focused on identifying or building a representative open science dataset that addresses biases in Alzheimer鈥檚 research. Among thousands of datasets, Lanzi鈥檚 DementiaBank emerged as the standout winner.
鈥淚f we really want to take a crack at early detection and make advancements, we need a team-based, collaborative approach鈥攕omething much larger than a single research lab at any university,鈥 Lanzi said. 鈥淭o see data scientists from all over the U.S. with industry backgrounds from Google and Amazon come together and use the data we鈥檙e collecting at 91原创 to build analytical approaches that will drive the early detection field could not be more motivating.鈥
DrivenData, which organized the challenge, called DementiaBank impressive.
鈥淲e were looking for that unicorn dataset,鈥 said Christine Chung, a senior data scientist with DrivenData. 鈥淒ementia Bank is a well-structured, accessible dataset with diverse representation. It鈥檚 an underexplored area, and machine learning and artificial intelligence now allow us to extract richer features from audio samples.鈥
Phase II, BuildIT! focused on advancing state-of-the-art, ethical, and inclusive algorithms and analytical approaches for early detection and prediction. Phase III, Put IT All Together! brought top teams together to demonstrate their models and pitch solutions.
DementiaBank was the sole database selected for acoustic analysis and was used throughout the challenge.
鈥淩ecruiting participants and collecting data in a standardized way is hard work,鈥 said Lanzi. 鈥淲e鈥檙e still in active data collection 鈥 and not even close to finished, so to see the impact of our work beyond publications, paving the way for leading scientific approaches, is incredible.鈥
Anna Saylor, a fourth-year student in the CSCD doctoral program, has been working in Lanzi鈥檚 since the project's inception. Having watched her late grandparents progress through Alzheimer鈥檚, the work feels deeply personal.
鈥淚 collected the very first data sample,鈥 Saylor said. 鈥淭o think about all the older adults who鈥檝e dedicated their time to this research and that these two-hour sessions can now be used for sophisticated approaches to early detection is mind-blowing.鈥
Her experience underscores the importance of understanding biomarkers in the voice.
鈥淒ementiaBank breaks down how vocabulary changes over time and introduces disfluencies, including stutters, pauses, and changes in pitch,鈥 Chung said. 鈥淚鈥檇 love to see this research evolve into a product that can help detect Alzheimer鈥檚 20 years earlier in a way that鈥檚 cost-effective and accessible for everyone.鈥
Current diagnostic tools for detection, like MRI, are costly and invasive, and Lanzi believes that language, while not a single predictor, could be a powerful piece of the puzzle.
鈥淚 hope language becomes a marker that bridges the gap and gets people involved in treatments as early as possible. That could change their trajectory or help them better manage their function as Alzheimer鈥檚 progresses,鈥 Lanzi said.
The challenge has also opened new doors for collaboration.
鈥淚鈥檝e had the chance to educate these innovators on mild cognitive impairment and give them a glimpse into the people behind the data, to humanize it,鈥 said Lanzi. 鈥淒ata scientists must understand that these are real humans 鈥 someone鈥檚 grandmother 鈥 behind the language they鈥檙e analyzing.鈥
Talking with data scientists has also helped Lanzi refine her data collection process.
聽鈥淭oo often, we operate in silos,鈥 said Lanzi. 鈥淏ringing everyone together to understand what they need, to learn how I can refine my processes, and explain why some elements can鈥檛 be changed is pivotal.鈥
Challenge organizers say 91原创鈥檚 contributions to open science are setting a national standard.
鈥淭here鈥檚 no question that 91原创 is a leader in open science and early detection efforts,鈥 said Chung. 鈥淒ementiaBank is a clear leader in this field 鈥 it鈥檚 a wealth of data, and I don鈥檛 think there鈥檚 another resource like it out there.鈥
This kind of crossover between industry and academia is rare; it鈥檚 a partnership Lanzi wants to see grow.聽聽
鈥淥ne day, I hope we have teams comprised of clinicians, patient-centered researchers, and data scientists,鈥 she said. 鈥淲hen the best of the best problem-solvers come together, anything is possible.鈥
Contact Us
Have a 91原创aily story idea?
Contact us at ocm@udel.edu
Members of the press
Contact us at mediarelations@udel.edu or visit the Media Relations website
