Researchers in the US reported that a deep-learning model trained on eye-tracking data could distinguish young adults with self-reported depression or suicidal ideation from healthy controls, suggesting eye movements may offer an objective behavioural marker of symptom severity.  

Writing in npj Digital Medicine, the team from the University of Southern California and UCLA analysed eye movements from 126 young adults as they read and responded to emotionally-loaded sentences such as ‘My mind gravitates toward thoughts of sadness’ and ‘My mind gravitates toward thoughts of joy’. 

The model was reasonably accurate at distinguishing participants with depression or suicidal ideation from healthy controls (area under the curve (AUC) 0.793) and performed slightly better for suicidal ideation alone (AUC 0.826). However, it was much less accurate at distinguishing depression from suicidal ideation (AUC 0.609). The clearest differences in eye movements appeared while participants were generating responses, especially when they were reading negative sentences, suggesting emotionally negative language may make mood-related eye-movement differences easier to detect.