Differentiating depression using facial expressions in a virtual avatar communication system.

Area of Science:

• Human-Computer Interaction
• Psychology
• Affective Computing

Background:

• Depression significantly impacts quality of life, necessitating effective detection methods.
• Human-machine interaction offers novel avenues for mental health assessment.
• Virtual avatar communication systems combined with facial expression monitoring show potential for depression classification.

Purpose of the Study:

• To assess the impact of human versus virtual avatar interviewers on individuals with depressive symptoms.
• To clarify how neutral conversation topics influence facial expressions and emotions in those with depression.
• To compare verbal and non-verbal cues between individuals with and without depression.

Main Methods:

• Twenty-seven participants (15 control, 12 depression symptoms) interacted with virtual avatars and human interviewers.
• Participants engaged in neutral and negative conversation topics, completing the PANAS scale.
• Facial expressions were recorded via webcam and analyzed using manual annotation and automatic detection (OpenFace).

Main Results:

• No significant differences in emotions were found between human and virtual avatar interviewers.
• Neutral topics induced less negative emotion compared to negative topics for both groups.
• Distinct facial expression patterns were observed between individuals with and without depression within the virtual avatar system.

Conclusions:

• Virtual avatars and human interviewers did not significantly differ in emotional or facial expression outcomes.
• Neutral conversation topics were associated with reduced negative emotions in both control and depression groups.
• The virtual avatar communication system demonstrated potential for identifying differing facial expression patterns indicative of depression.