Multimodal physiological responses evoked by emotional naturalistic stimuli facilitate classification of major depression and generalized anxiety.

The diagnosis of major depressive disorder (MDD) and generalized anxiety disorder (GAD) poses a substantial burden to society. Identifying effective biomarkers for these disorders is essential for early diagnosis and alleviating clinical burden. However, existing laboratory-based paradigms and neuroimaging-based frameworks often suffer from limited ecological validity and high resource demands, thereby restricting their applicability in large-scale, real-world settings. Here, we propose a classification framework that combines multimodal physiological signal recordings-including eye-tracking, heart rate variability (HRV), and electrodermal activity (EDA) data-together with an emotional naturalistic stimuli paradigm designed to elicit rich and dynamic physiological responses. Our results indicate that the emotional naturalistic stimuli effectively elicit diverse physiological responses across MDD and GAD patients. Next, the early fusion of 3 physiological modalities achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.87 for classifying patients from healthy controls (HC) (task 1), and an AUROC of 0.79 for differentiating between MDD and GAD (task 2). Further post-hoc feature contribution analyses revealed that the patient-HC classification relied on evenly distributed contributions from three physiological modalities, with the top 5 features spanning four emotional stimuli, whereas MDD-GAD differentiation was dominately contributed by EDA responses to fear stimulus. Taken together, these results suggest that integration of multimodal physiological responses combined with the emotional naturalistic stimuli paradigm supports the development of portable, low-burden adjunctive frameworks for classifying MDD, GAD, and the HC group.