A Non-Invasive 3D Gait Analysis Framework for Quantifying Psychomotor Retardation in Major Depressive Disorder

TL;DR

This work tackles the challenge of objective psychomotor retardation assessment in Major Depressive Disorder by introducing a non-invasive, monocular RGB gait analysis framework. It converts video into 3D gait kinematics using Gravity-View Coordinates with a trajectory correction to mitigate monocular drift, yielding 297 explicit digital biomarkers. On the CALYPSO dataset (N=42), the method achieved an $Acc=83.3\%$ for PMR classification and an $R^2=0.64$ for depression severity, with interpretable signatures such as reduced ankle propulsion and restricted pelvic mobility underpinning the predictions. The approach offers a scalable, transparent tool for routine clinical monitoring of depression, bridging gait analysis with psychiatric semiology and providing stable signatures suitable for longitudinal deployment.

Abstract

Predicting the status of Major Depressive Disorder (MDD) from objective, non-invasive methods is an active research field. Yet, extracting automatically objective, interpretable features for a detailed analysis of the patient state remains largely unexplored. Among MDD's symptoms, Psychomotor retardation (PMR) is a core item, yet its clinical assessment remains largely subjective. While 3D motion capture offers an objective alternative, its reliance on specialized hardware often precludes routine clinical use. In this paper, we propose a non-invasive computational framework that transforms monocular RGB video into clinically relevant 3D gait kinematics. Our pipeline uses Gravity-View Coordinates along with a novel trajectory-correction algorithm that leverages the closed-loop topology of our adapted Timed Up and Go (TUG) protocol to mitigate monocular depth errors. This novel pipeline enables the extraction of 297 explicit gait biomechanical biomarkers from a single camera capture. To address the challenges of small clinical datasets, we introduce a stability-based machine learning framework that identifies robust motor signatures while preventing overfitting. Validated on the CALYPSO dataset, our method achieves an 83.3% accuracy in detecting PMR and explains 64% of the variance in overall depression severity (R^2=0.64). Notably, our study reveals a strong link between reduced ankle propulsion and restricted pelvic mobility to the depressive motor phenotype. These results demonstrate that physical movement serves as a robust proxy for the cognitive state, offering a transparent and scalable tool for the objective monitoring of depression in standard clinical environments.

Figures (6)

• Figure 1: Overview of the proposed pipeline. Monocular video is processed through GVHMR to recover SMPL body parameters: shape ($\beta$), pose ($\theta$), global orientation ($\phi$), and translation ($t$). Trajectory refinement (loop closure, PCA alignment) corrects $\phi$ and $t$ while preserving $\beta$ and $\theta$, producing coherent 3D skeleton sequences. Gait event detection identifies walking phases and arm swing cycles. Extracted features undergo stability-based selection for PMR classification or severity regression.
• Figure 2: Experimental setup. Top: Camera view of the clinical recording environment. Videos were captured using a Samsung Galaxy S21 FE smartphone (Android 13) at $1440\times1440$ resolution (1:1 aspect ratio) and 30 fps using the ultrawide lens. The device was stabilized on a tripod at a height of approximately 1.5 m. Bottom: Diagram of the modified Timed Up and Go (TUG) protocol, illustrating the four linear movement phases separated by turning segments, designed to ensure that the full TUG trajectory remains within the camera frame.
• Figure 3: Turn detection during the TUG test. Top: Progression along the principal walking axis with detected turn centers (red diamonds). Bottom: Velocity profile showing raw (light blue) and smoothed (dark blue) signals. Turns are identified when velocity falls below the threshold (dashed red line), with turn regions shaded.
• Figure 4: Gait event detection during the TUG protocol. Left foot heel (solid) and toe (dashed) trajectories relative to the pelvis are shown. Green upward triangles indicate Heel Strikes (HS); purple downward triangles denote Toe-Offs (TO). Background colors distinguish between protocol phases: blue (outward walking), red (turning), and green (returning).
• Figure 5: Ankle dorsiflexion across normalized gait cycles for a representative participant. Shaded regions indicate standard deviation across $n=3$ cycles.