USC Researchers Use Sweat, Brain Signals, and Eye Tracking to Detect Depression

Objective biomarkers for mental health assessment

Researchers at the University of Southern California have developed a multimodal system that uses biological signals—sweat responses, brain activity, and eye movements—to identify depression and suicidal ideation. The approach, funded by the Defense Advanced Research Projects Agency (DARPA), aims to provide psychiatrists with objective data to supplement traditional self-report surveys and clinical interviews.

The PRECOG (Multimodal integration of neural and biobehavioral signals for predicting preconscious responses) project, led by Shrikanth Narayanan, USC's inaugural vice president for presidential initiatives and University Professor, has produced five published papers in major journals since the research began in June 2023. The work brings together expertise from engineering, neurology, speech technology, artificial intelligence, linguistics, and clinical psychiatry.

Why it matters

Suicide rates in the United States have climbed steadily for decades, with veterans experiencing rates 1.5 times higher than the general population, according to USC research. Many at-risk individuals never seek help due to stigma and reporting biases. Because mental health diagnoses still rely heavily on patient self-reporting, clinicians lack standardized testing methods comparable to those used for physical illnesses. An objective, data-driven assessment tool could help identify high-risk individuals earlier and enable more timely interventions.

How the research worked

The study centered on a sentence evaluation task in which participants read 160 self-referential statements with varying emotional tones, such as "I feel sad a lot of the time." While participants indicated agreement or disagreement with each statement, researchers simultaneously recorded brain activity using a 64-channel electroencephalography (EEG) system, tracked eye movements with an infrared SR Research EyeLink 1000 Plus system, and measured skin conductance responses through electrodermal activity (EDA) sensors.

Researchers then applied deep learning techniques to identify patterns distinguishing healthy individuals from those with depression or suicidal ideation. The team found that neural responses differed most reliably 300 to 600 milliseconds after word presentation, a window associated with emotional and semantic processing. PhD student Woojae Jeong's analysis revealed that stronger differential responses to positive versus negative sentences helped distinguish healthy individuals from those with depression.

In eye-tracking analysis led by PhD student Kleanthis Avramidis, horizontal gaze patterns carried the most diagnostic information, particularly when participants processed negative statements. Healthy participants showed more structured visual patterns aligned with the "disagree" option for negative content, while individuals with suicidal ideation exhibited more dispersed or disengaged viewing patterns.

The sweat response analysis found that reactions to negative words carried the strongest diagnostic information. Individuals with depression and suicidal ideation showed altered physiological reactions to emotionally distressing content, suggesting that sympathetic nervous system reactivity could serve as a measurable biomarker.

A supplemental tool, not a replacement

The researchers emphasize their work is designed to add an additional layer to existing assessments rather than replace clinician judgment. "Our goal isn't to replace current mental health assessments but to supplement them," said Dani Byrd, a professor at USC Dornsife's department of linguistics and co-author on the papers. The tool could help psychiatrists make more informed, data-driven decisions and identify more precise intervention points for patients experiencing suicidal ideation.

The details were first reported by USC Viterbi School of Engineering.