Associations of AI-Based Facial Metrics With Patient-Reported Outcomes in Idiopathic Facial Paralysis.

Abstract

Conventional grading scales and static image assessments may not capture dynamic facial movement in facial paralysis. We developed a video-based, dynamic, artificial intelligence (AI) application, DynaFace, to objectively quantify facial metrics and determine which measures correspond with patient-reported appearance, function, and psychosocial outcomes, providing insight into how objective dynamics relate to subjective patient experience. DynaFace automatically extracted dynamic facial metrics (facial asymmetry index [FAI], bilateral palpebral fissure, and oral commissure excursion [OCE]) at rest and during smile, laughter, and pucker. Multivariable linear regression, controlling for paralysis duration and baseline trait emotional intelligence, assessed associations between objective metrics and patient-reported outcomes from FACE-Q subsets. Greater asymmetry (higher FAI) during smile and laugh was associated with poorer perceived facial (β = smile: -2.82/laugh: -3.01, p < 0.05), eye (β = -3.99/-3.12, p < 0.05), and smile appearance (β = -4.40/-4.80, p < 0.01), as well as lower overall facial function (β = -3.05/-3.60, p < 0.05) and self-esteem (β = -2.93/-3.12, p < 0.05). In contrast, greater ratio of OCE change (affected to unaffected hemiface) during smile predicted better perceived facial appearance (β = 0.88, p < 0.01), smile appearance (β = 0.69, p < 0.05), and higher self-esteem (β = 0.62, p < 0.05). Greater ratio of OCE change during both smile and laugh predicted improved social function (β = 0.48/0.44, p < 0.05). Palpebral fissure asymmetry was associated with only eye appearance. Dynamic AI-derived facial metrics from DynaFace align with patient perceptions, showing that greater symmetry in movement enhance satisfaction and function. These findings highlight the clinical potential of DynaFace to bridge objective and patient-reported measures in facial paralysis assessment.