Using this new form of artificial intelligence, developed by the University of California and Northwestern University, the technology was able to predict whether a respondent had high or low anxiety levels with up to 81% accuracy. The system also scored highly on sensitivity and specificity, performance measures that show how well a model classifies people with high or low anxiety levels.
“Picture assessment tasks can be used to create a day-to-day, unbiased snapshot of a person's mental health status without asking direct questions that may evoke negative or upsetting emotions,” Bali noted.
She added that direct questions about anxiety may become less effective over time because respondents tend to end up answering the same questions mechanically. “Because this new technique is not native language dependent, it can be used broadly in different settings to assess anxiety.”
The researchers noted that the anonymous data was drawn from a U.S. population during the COVID-19 pandemic, which has seen people report higher levels of loneliness and anxiety than usual.
The research was funded by two grants awarded to Breiter from the U.S. Office of Naval Research. Breiter, Bari, Nicole Baik, and engineer Byung-woo Kim collaborated with researchers including Shamal Lalvani, Leandros Stephanopoulos, Martin Block, and Katsageros (co-senior author) at Northwestern University, and Nikos Magravelas at Aristotle University of Thessaloniki.
Massachusetts General Hospital (Harvard Medical School) is also actively involved in this research, and the research team led by Breiter was the first to discover the computational cognition that underlies this work.
Featured image at top: Senior research scientist Sumra Bali is researching new health applications of artificial intelligence in the UC School of Engineering and Applied Science. Photo/Andrew Higley/UC Marketing + Branding
