UNIVERSITY PARK, Pa. — Over 7 million people aged 65 and older suffer from Alzheimer's disease in the United States, according to a 2025 report from the Alzheimer’s Association. More of the debilitating symptoms could be mitigated or better managed with an earlier diagnosis, said Hui Yang, Gary and Sheila Bello Chair in Industrial and Manufacturing Engineering at Penn State.
Yang, alongside Kevin Mekulu, an industrial engineering and operations research doctoral candidate, recently co-authored a series of papers exploring how artificial intelligence (AI) could be used to better understand neurodegenerative diseases like dementia and Alzheimer's. Their work, which was published in the Journal of Alzheimer's Disease Reports and Frontiers in Aging Neuroscience, proposes a novel framework of detecting possible problems based on patterns found in transcriptions of patient speech. The new approach could detect dementia earlier and with more accuracy than the paper-based exams traditionally used, the researchers suggest.
In the following Q&A, Yang and Mekulu shared how AI-powered screening methods could contribute to cognitive care, potentially helping clinicians intervene sooner and improve patient outcomes.
Q: What benefits does AI offer over traditional screening approaches for dementia and Alzheimer's?
Yang: Traditional dementia screening tools are paper-based, subjective and resource-intensive, requiring 10 to 15 minutes of staff time for administration, while lacking sensitivity to subtle cognitive changes and showing inconsistency between proctors. With the U.S. facing a shortage of geriatric specialists, having roughly one geriatrician for every 10,000 geriatric patients and high staff turnover in senior care facilities, a scalable AI solution is urgently needed. Our framework uses interpretable, speech-based biomarkers to capture subtle linguistic changes and cognitive decline years before traditional tools can, offering objective and non-invasive screening for neurodegenerative conditions in under a minute.
Q: What makes your approach different from the “static” AI models already employed in some screening approaches?
Mekulu: Most AI models used today in health care are static, meaning they simply produce an output based on an input. Agentic AI, by contrast, are systems capable of independently planning and executing complex tasks without human oversight. It is designed to reason over time, adapt its behavior and interact dynamically with patients or clinicians. In our work, AI agents are not just scoring a test — they guide a screening interaction, adapt prompts based on a person’s responses and integrate multiple signals, such as language patterns, task performance and contextual factors into a coherent assessment. This transforms screening from a one-time measurement to an evolving process that better reflects how cognitive decline occurs in patients.
Q: One paper shows how speech patterns can be used as a tool for diagnoses. Why look at speech patterns? How does your AI do this?
Yang: Speech is one of the most information-dense behaviors humans produce, requiring the coordination of memory, attention, language, executive function and motor planning — all cognitive systems that are affected early in neurodegenerative disease.
Our AI analyzes complex dynamics and transitions hidden in speech rather than relying on subjective clinical impressions alone, searching for subtle patterns in word choice, repetition, fluency changes and the structural organization of language to reveal cognitive changes long before symptoms become obvious. This approach allows us to extract objective, quantitative biomarkers from natural patient behavior, which removes a lot of the subjective interpretation associated with traditional tests.
Q: Are there other activities or behaviors AI could analyze to detect neurodegenerative disease? Could AI agents be implemented into other aspects of treatment?
Mekulu: Absolutely. Speech is a powerful starting point, but it’s only one piece of the puzzle. We can use AI to analyze eye-movement patterns, physiological signals, task engagement, motor behavior and even how someone learns or adapts over time during problem-solving tasks. Interpreting all these signals together offers clinicians a more holistic view of cognitive health, not just whether someone passes or fails a test.
AI agents could eventually support care planning, monitor cognitive changes between clinic visits, and help clinicians identify when interventions need to be adjusted. Rather than replacing clinicians, these systems are designed to reduce administrative burden, highlight meaningful patterns and help transform cognitive care from reactive to preventative.
Q: What’s next for this work?
Yang: We are actively evaluating these methods across different populations and clinical contexts to ensure robustness and fairness. Additionally, we are working with Dr. Nicole Etter, associate professor in the Department of Communication Sciences and Disorders at Penn State, and Dr. Tim Brearly, a neuropsychologist from Penn State Health, to see how these tools can be integrated into assisted living and memory care environments in ways that are practical for patients and clinicians. These settings are often where subtle cognitive changes first emerge, yet objective screening tools are rarely used at scale. We aim to bridge the gap between academic research and everyday clinical decision-making by validating these methods in real-world care environments.