Multi-source Information Fusion for Depression Detection

Depression is the most common psychiatric disorder. Traditional depression detection methods almost rely on structured scales and clinical opinions, which carry the risk of subjective judgment. In light of this, we investigate the potential of employing emotional images as stimuli for depression detection. Our proposed method is the first to utilize pupil dilation, blink patterns, and eye movements as features for depression detection. Notably, we introduce a comprehensive set of strategies for extracting visual cognitive features, validating the efficacy of the pupil emotion response theory and blink emotion response theory. Finally, we train a Support Vector Machine (SVM) classifier to differentiate between depressed and normal subjects, achieving an impressive accuracy of 89.5%, which is higher than other state-of-the-art methods in automatic depression detection.