Detecting Depression Severity by Interpretable Representations of Motion Dynamics.

Area of Science:

• Computational psychiatry
• Machine learning in healthcare
• Biomedical signal processing

Background:

• Deep learning enables objective depression severity measurement via facial and head motion analysis.
• Current deep learning methods lack interpretability, hindering clinical application and understanding.
• Need for interpretable models in mental health diagnostics.

Purpose of the Study:

• To develop an interpretable method for automatic depression severity measurement.
• To utilize facial landmarks and 3D head motion for kinematic feature extraction.
• To classify depression severity levels using machine learning models.

Main Methods:

• Employed barycentric coordinates of facial landmarks and Lie-algebra based rotation matrices for 3D head motion.
• Extracted, preprocessed, and encoded kinematic features using Gaussian Mixture Models (GMM) and Fisher vector encoding.
• Utilized a multi-class Support Vector Machine (SVM) for classifying depression severity.

Main Results:

• The interpretable method achieved classification accuracy comparable to state-of-the-art deep learning approaches.
• Facial movement velocity and acceleration significantly correlated with depression severity.
• Findings align with established clinical data and theories of depression symptomatology.

Conclusions:

• An interpretable machine learning framework can objectively measure depression severity using facial and head kinematics.
• The method provides clinically and theoretically relevant insights into depression.
• This approach offers a promising tool for objective mental health assessment.