Automatic Identification of a Depressive State in Primary Care

Area of Science:

• Computational psychiatry
• Affective computing
• Digital mental health

Background:

• The Center for Epidemiologic Studies Depression Scale (CES-D) is effective but lengthy for primary care.
• There is a need for brief, objective depression screening tools.
• Social media platforms offer opportunities for ecological data capture via facial movement.

Purpose of the Study:

• To establish an automatic depression recognition model using facial and gaze behaviors.
• To develop a tool for accessible primary healthcare screening.
• To compare different machine learning algorithms and features for optimal performance.

Main Methods:

• Integrated facial activities and gaze behaviors into a machine learning algorithm (Kernel Ridge Regression, KRR).
• Compared various algorithms and feature sets to identify the best performing model.
• Utilized nonverbal behavioral cues associated with depressive states.

Main Results:

• The model incorporating both facial and gaze features outperformed models using only facial features.
• The ridge model with a periodic kernel demonstrated the best performance.
• Achieved a cross-validated R-squared (R2) of 0.43 and a Pearson correlation coefficient (r) of 0.69 (p < 0.001).

Conclusions:

• Facial and gaze behavior analysis provides a viable method for automatic depression recognition.
• The developed KRR model shows potential for integration into primary healthcare settings.
• Key variables include specific gaze directions and facial action units relevant to depression.