Measuring Anxiety Levels with Head Motion Patterns in Severe Depression Population

TL;DR

This work tackles the challenge of noninvasively assessing anxiety in individuals with severe depression by analyzing head motion during video interviews. It introduces the CALYPSO Depression Dataset and a pipeline that extracts head pose and motion, segments interviews into moving and steady phases via a Gaussian Mixture Model, derives 283 features, and selects robust predictors to regress psychological anxiety levels with a linear model regularized by Lasso. The best model achieves MAE = 0.31 and R^2 = 0.87 using 14 features, demonstrating that structured head-motion cues can serve as objective indicators of anxiety severity in depression. The approach offers a practical, interpretable tool for clinicians, enabling rapid, noninvasive assessment and potential personalization of treatment strategies, with clear pathways for multimodal and longitudinal extensions.

Abstract

Depression and anxiety are prevalent mental health disorders that frequently cooccur, with anxiety significantly influencing both the manifestation and treatment of depression. An accurate assessment of anxiety levels in individuals with depression is crucial to develop effective and personalized treatment plans. This study proposes a new noninvasive method for quantifying anxiety severity by analyzing head movements -- specifically speed, acceleration, and angular displacement -- during video-recorded interviews with patients suffering from severe depression. Using data from a new CALYPSO Depression Dataset, we extracted head motion characteristics and applied regression analysis to predict clinically evaluated anxiety levels. Our results demonstrate a high level of precision, achieving a mean absolute error (MAE) of 0.35 in predicting the severity of psychological anxiety based on head movement patterns. This indicates that our approach can enhance the understanding of anxiety's role in depression and assist psychiatrists in refining treatment strategies for individuals.

Figures (13)

• Figure 1: Overview of our proposed pipeline. We apply the same process to all videos of the informal interviews of the CALYPSO dataset. We first extract the head pose and its motion (speed and acceleration) automatically. We then apply statistical feature analysis to extract a feature vector of size 283 for each video. Finally, we apply a cross-validated approach to select features and train a linear model to accurately regress psychological anxiety levels.
• Figure 2: Interview Room Setup for the Calypso Depression Dataset.
• Figure 3: Diagram of head motion axes—pitch, roll, and yaw—used in our analysis.
• Figure 4: Illustration of the head pose and motion extraction.
• Figure 5: We apply a Gaussian Mixture Model (GMM) to cluster head rotational velocities (pitch, yaw, roll) into "moving" and "steady" states, segmenting interviews into sequences (e.g., F21-F60, representing frames 21 to 60). The plot illustrates the velocity profiles for each axis.