Researchers at UC Irvine have moved beyond the lab, installing AI-powered monitoring systems in actual dementia care facilities to detect emotional distress and prevent dangerous behavioral episodes in real time. The interdisciplinary project, led by Distinguished Professor Adey Nyamathi and supported by Amir Rahmani, professor of nursing and computer science, recently completed its initial pilot phase and is preparing to submit a larger National Institutes of Health (NIH) proposal to expand the work.

The breakthrough moment came when researchers began collecting data directly from dementia patients in residential board-and-care facilities, moving from theoretical models to real-world validation. This shift from controlled environments to actual care settings represents a critical milestone in developing technology that can genuinely help vulnerable populations.

What Makes This AI System Different from Standard Monitoring?

The system deployed in patient rooms goes far beyond simple video surveillance. Researchers equipped the spaces with a sophisticated multimodal monitoring setup that captures multiple types of data simultaneously, allowing the AI to understand what's happening from multiple angles.

• Audio Analysis: The system examines vocal tone, speech frequency, and communication patterns that may signal agitation or emotional changes, not just the words being spoken.
• Visual Monitoring: Cameras track movement patterns and gait changes that can indicate distress or behavioral shifts in dementia patients.
• Physiological Sensors: Heart rate, respiration, and other vital signs provide real-time data about a patient's physical state during moments of potential crisis.

"We started collecting data in real dementia patients. That's been the biggest milestone," said Amir Rahmani.

Amir Rahmani, Professor of Nursing and Computer Science at UC Irvine

The AI models evaluate all this information together to recognize emotional states and behavioral patterns. Researchers also analyze interactions between caregivers and patients to identify specific cues linked to emotional distress, creating a more nuanced understanding of what triggers agitation.

Why Does Detecting Agitation Matter So Much in Dementia Care?

Agitation is far more than an inconvenience in dementia care; it's a major safety risk. The condition is one of the most common and dangerous behavioral symptoms associated with dementia, and it directly contributes to falls and other serious complications.

"Agitation is one of the major causes of falls and other complications. If we can recognize when someone is becoming distressed and help calm them down, even temporarily, it can give caregivers time to intervene and potentially prevent catastrophic events," explained Amir Rahmani.

Amir Rahmani, Professor of Nursing and Computer Science at UC Irvine

Early findings from the pilot study have revealed a concerning pattern that researchers didn't initially expect. Many dementia patients experienced repeated sleep disruptions throughout the night, and these patterns appeared directly connected to increased agitation and mood changes the following day. This discovery suggests that addressing sleep quality could be a powerful intervention point for reducing behavioral crises.

"We could see patterns emerge from both the camera data and the physiological data. Sleep disturbances appear to affect mood, agitation and overall condition the next day," noted Amir Rahmani.

How to Implement Privacy-First AI in Healthcare Settings

One of the most critical aspects of this project often gets overlooked: the infrastructure required to protect patient privacy when collecting highly sensitive audio, video, and health data. The research team spent months building secure systems that prioritize data protection.

• Internal Server Infrastructure: Rather than relying on third-party cloud providers, researchers developed their own internal server systems to maintain complete control over sensitive patient data.
• Access Restrictions: Only the research team can access the collected data, with strict protocols preventing unauthorized viewing or analysis.
• Secure Data Collection: The system was designed from the ground up with privacy as a core requirement, not an afterthought added to existing technology.

"A lot of effort went into making sure everything is private and secure. Only the research team can access the data," stated Amir Rahmani.

Amir Rahmani, Professor of Nursing and Computer Science at UC Irvine

What's Next for AI-Powered Dementia Care?

The current phase of the project focuses on detection and pattern recognition, but the long-term vision is far more ambitious. Researchers plan to introduce generative AI capabilities that would allow a robotic companion to engage patients in calming conversations modeled after successful caregiver interactions.

The idea is for the system to eventually learn from real caregiving scenarios, including personal stories and conversational techniques that have proven effective at reducing distress in dementia patients. However, this portion of the project is still in development because researchers need significantly larger datasets before training advanced generative AI models.

The team is now seeking a multi-year NIH grant that would allow them to scale the project, validate the technology across more facilities, and continue developing intervention tools designed to support both patients and caregivers. While widespread use is still years away, the progress made so far demonstrates how AI-assisted technologies can work in real environments with real patients, marking a critical step forward in improving quality of life for dementia patients and easing caregiver burden.