Emotional processing deficits are increasingly recognized as a salient feature of psychotic disorders, and emerging evidence suggests that impairments in facial emotion processing may serve as additive predictive markers for individuals at ultra-high risk (UHR) for psychosis, highlighting the importance of such deficits for early identification and intervention strategies. In this study we examined whether automated coding of facial features from brief video recordings (45 seconds) could discriminate 108 individuals at ultra-high risk for psychosis (UHR) from 65 matched healthy controls (HC). We assessed the alignment between automated coding of facial expressivity as measured by granulated muscle activity, and clinician-rated facial emotional expressivity by experienced clinicians. In addition, we explored associations with social skills and clinical symptoms. A random forest classifier with repeated nested cross-validation significantly classified UHR-individuals from HC with above chance accuracy (63%; p = 0.009, sensitivity=0.76, specificity=0.40). Automated coding of Presence, Intensity, and Variation of specific facial features correlated strongly with clinician-rated facial expressivity with moderate to large effect sizes (all corrected p 0.001). Moreover, facial feature metrics were associated with attenuated psychotic symptoms and negative symptoms, supporting the clinical validity of automated facial expressivity coding from brief video recordings of UHR-individuals.
Kristensen et al. (Wed,) studied this question.