Machine learning classification of schizophrenia and social anhedonia based on facial expressions.

BACKGROUND AND HYPOTHESIS: Schizophrenia (SCZ) is characterized by deficits in emotional expression, with facial expressions serving as potential markers for diagnosis. Leveraging machine learning and computerized facial analysis, this study aimed to identify facial expression features in patients with SCZ and individuals with high social anhedonia (SocAnh), construct an explainable classification model, and explore the associations between facial features and negative symptoms. STUDY DESIGN: Emotional expressions of 2 samples comprising 32 patients with SCZ and 34 Health controls (HC), and 56 participants with high SocAnh and 56 participants with low SocAnh were recorded based on an emotion elicitation paradigm combining film based and autobiographical methods. Facial features were extracted using FaceReader to develop classification models with a standardized machine learning pipeline. STUDY RESULTS: Patients with SCZ showed lower intensity of sad facial expressions during neutral and positive film viewing than HC, whereas no significant differences were found between participants with high and low SocAnh. In the clinical sample, the Support Vector Machine model achieved a classification accuracy of 81.7%, with highly weighted features showing a mixed pattern across elicitation conditions. In contrast, the model distinguishing participants with high versus low SocAnh yielded lower classification performance. Exploratory correlation analyses showed modest associations between selected facial expression features and negative symptoms, but none remained statistically significant after Bonferroni correction. CONCLUSIONS: Computational facial expression analysis with interpretable machine learning may help examine facial expression deficits in SCZ. Associations with negative symptoms were exploratory and require confirmation in larger independent samples.