Artificial Intelligence-Derived Emotionality State Score Correlates with Layperson Assessment and Objective Measurements in Facial Paralysis.

BACKGROUND: There is growing interest in using artificial intelligence (AI) to track the outcomes of patients with facial paralysis (FP). However, validation studies of the AI output with other outcome measurements are still scarce in the literature. We aimed to correlate an AI-derived emotionality state score (ESS) with layperson assessment and objective oral commissure excursion in facial reanimation outcomes. METHODS: In this retrospective cohort study, videos of voluntary smiles from 63 patients with FP across a variety of facial reanimation procedures were analyzed by the FaceReader software before and after the surgery. The ESS was defined as the intensity score (IS) of happiness emotion minus the IS of the negative emotion (sadness, fear, anger, disgust, contempt) with the highest IS. Five lay observers rated the videos with the Terzis score. The oral commissure excursion was measured with the Emotrics software. RESULTS: For smiling with teeth showing, there was a significant (p<0.05) postoperative improvement in the ESS, as well as a significant decrease in the median IS of all negative emotions. The correlation of the ESS was strong with the Terzis score (r=0.78, p<0.001) and moderate with the oral commissure excursion (r=0.59, p<0.001). The software showed better performance in patients with complete paralysis compared to incomplete paralysis. CONCLUSIONS: The AI-derived emotionality recognition approach holds great promise for both automating layperson assessments and contributing to the objective evaluation of facial reanimation outcomes. Future studies should focus on training the AI on patients with FP and multicenter validation studies.