Multimodal AI System Tracks Early Warning Signs of Mental Health Crisis

In the United States, more than 47,000 people die by suicide each year and more than 10 million seriously consider it, according to the U.S. Centers for Disease Control and Prevention. Yet clinicians often have limited ways to identify when someone may be moving toward crisis before it occurs. Proactive, continuous risk assessment could support earlier and more targeted care for young adults. Building on this need, researchers are developing an AI platform to predict depression, suicide risk, and resilience in college students.

Researchers at the Keck School of Medicine of the University of Southern California (USC), with colleagues from the USC Viterbi School of Engineering and USC Dornsife College of Letters, Arts and Sciences, are building the Scalable Evaluation of Neurobehavioral Trajectories in Everyday Life (SENTINEL) platform. The system will analyze multimodal signals from brain activity and daily behavior to characterize risk and protective factors. The work is supported by the Advanced Research Projects Agency for Health (ARPA-H) with a contract of up to $4 million that may be renewed after two years for a total of up to $7 million.

The observational study will follow approximately 210 USC students for up to 36 months. Participants will be evenly divided among three groups: no mental health concerns, depression, and depression with suicidal thoughts or behaviors. Every six months, enrollees will complete laboratory assessments that include electroencephalography (EEG), eye tracking, and detailed psychiatric questionnaires.

Continuous monitoring outside the lab will pair wearable sensors with smartphone-based reporting. Devices such as an Oura Ring or Fitbit will record heart rate, heart rate variability, skin temperature, skin conductance, blood oxygen levels, physical activity, and sleep stages and duration. Weekly app-based questionnaires will capture stress, anxiety, mood, and daily experiences, with voice responses analyzed for speech patterns; with consent, smartphone data on app use, communication patterns, and attention will also be collected.

The protocol is designed to detect both periods of distress and periods of well-being to illuminate resilience. Earlier work from the team found that adults experiencing suicidal thoughts show distinct patterns in eye movements, brain activity, and other measurable biological signals such as sweat. The project is part of ARPA-H’s EVIDENT initiative, which aggregates de-identified, real-time clinical and digital data into a secure national repository to help identify rapid changes in mental health.

Although observational, the study aims to translate findings into tools that alert clinicians to early warning signs and inform timely intervention. Patterns in sleep, activity, or stress could help guide decisions about therapy, mindfulness exercises, medication, or other support. Over time, the technology could be adapted to monitor additional mental health conditions and to support high-stress groups such as health care workers, first responders, and military personnel.

Related Links
Keck School of Medicine of USC