Stony Brook Researchers Use AI to Advance Alzheimer’s Detection

Shan Lin

Alzheimer's disease is one of the most urgent public health challenges for aging Americans. Nearly seven million Americans over the age of 65 are currently living with the disease, and that number is projected to nearly double by 2060, according to the Alzheimer's Association.

Early diagnosis and continuous monitoring are crucial to improving care and extending independence, but there isn't enough high-quality, Alzheimer's-specific data to train artificial intelligence systems that could help detect and track the disease.

Shan Lin, associate professor of Electrical and Computer Engineeringat Stony Brook University, along with PhD candidate Heming Fu, are working with Guoliang Xing from The Chinese University of Hong Kong to create a network of data based on Alzheimer's patients. Together they developed SHADE-AD(Synthesizing Human Activity Datasets Embedded with AD features), a generative AI framework designed to create synthetic, realistic data that reflects the motor behaviors of Alzheimer's patients.

This figure provides the design overview of Shade-AD. The Training Process involves three stages: Stage 1 learns general human actions; Stage 2 embeds AD-specific knowledge; and Stage 3 fine-tunes the model based on patient-specific motion metrics.

Movements like stooped posture, reliance on armrests when standing from sitting, or slowed gait may appear subtle, but can be early indicators of the disease. By identifying and replicating these patterns, SHADE-AD provides researchers and physicians with the data required to improve monitoring and diagnosis.

Unlike existing generative models, which often rely on and output generic datasets drawn from healthy individuals, SHADE-AD was trained to embed Alzheimer's-specific traits. The system generates three-dimensional "skeleton videos," simplified figures that preserve details of joint motion. These 3D skeleton datasets were validated against real-world patient data, with the model proving capable of reproducing the subtle changes in speed, angle, and range of motion that distinguish Alzheimer's behaviors from those of healthy older adults.

The results and findings, published and presented at the 23rd ACM Conference on Embedded Networked Sensor Systems (SenSys 2025),have been significant. Activity recognition systems trained with SHADE-AD's data achieved higher accuracy across all major tasks compared with systems trained on traditional data augmentation or general open datasets. In particular, SHADE-AD excelled at recognizing actions like walking and standing up, which often reveal the earliest signs of decline for Alzheimer's patients.

This figure illustrates the comparison of "standing up from a chair" motion between a healthy elder and an AD patient.

Lin believes this work could have a significant impact on the daily lives of older adults and their families. Technologies built on SHADE-AD could one day allow doctors to detect Alzheimer's sooner, track disease progression more accurately, and intervene earlier with treatments and support. "If we can provide tools that spot these changes before they become severe, patients will have more options, and families will have more time to plan," he said.

With September recognized nationally as Healthy Aging Month, Lin sees this research as part of an effort to use technology to support older adults in living longer, healthier, and more independent lives. "Healthy aging isn't only about treating illness, but also about creating systems that allow people to thrive as they grow older," he said. "AI can be a powerful ally in that mission."

- Beth Squire