Cross-Cultural Detection of Depression from Nonverbal Behaviour

Area of Science:

• Psychiatry
• Computer Science
• Human-Computer Interaction

Background:

• Depression affects millions globally, necessitating accessible screening tools.
• Nonverbal behaviors may offer cross-culturally valid indicators of depression severity.
• Current depression assessment methods often lack cross-cultural applicability.

Purpose of the Study:

• To investigate the generalizability of a nonverbal behavior analysis approach for detecting depression severity across different cultures.
• To evaluate the impact of training data diversity on the cross-cultural performance of depression detection models.

Main Methods:

• Analysis of temporal features of eye gaze and head pose from video-recorded clinical interviews.
• Utilized datasets from Australia, the USA, and Germany, encompassing diverse interview types, depression subtypes, and healthy controls.
• Employed various training and testing strategies, including leave-one-subject-out cross-validation across combined datasets.

Main Results:

• The strongest cross-cultural depression severity detection performance was achieved when models were trained on data from all three countries.
• Generalizability decreased significantly when models were trained on data from only one or two countries and tested on unseen cultural data.
• Leave-one-subject-out cross-validation across all datasets demonstrated robust performance.

Conclusions:

• Training machine learning models on diverse, multi-cultural datasets is crucial for developing generalizable depression detection systems.
• Temporal features of nonverbal behaviors like eye gaze and head pose show potential for cross-cultural depression assessment.
• Future research should focus on incorporating a wide range of variability in training data to enhance the real-world applicability of automated mental health tools.