Automated Spontaneity Assessment after Smile Reanimation: A Machine Learning Approach

Background: Recreation of a spontaneous, emotional smile remains a paramount goal of smile reanimation surgery. However, optimal techniques to reliably restore spontaneity remain unknown. Dual automated machine-learning tools were used to develop an objective tool to analyze spontaneous smiling. The feasibility of this tool was tested in a sample of functional free muscle transfers. Methods: Validated humorous videos were used to elicit spontaneous smiles. Automated facial landmark recognition (Emotrics) and emotion detection software (Affdex) were used to analyze video clips of spontaneous smiling in nine normal subjects and 39 facial reanimation cases. Emotionality quotient was used to quantify the ability of spontaneous smiles to express joy. Results: The software could analyze spontaneous smiling in all subjects. Spontaneous smiles of normal subjects exhibited median 100 percent joy and 0 percent negative emotion (emotional quotient score, +100/0). Spontaneous smiles of facial palsy patients after smile reanimation, using cross-facial nerve graft, masseteric nerve, and dual innervation, yielded median emotional quotient scores of +82/0, 0/-48, and +10/-24 respectively (joy, p = 0.006; negative emotion, p = 0.034). Conclusions: Computer vision software can objectively quantify spontaneous smiling outcomes. Of the retrospective sample of cases reviewed in this study, cross-facial nerve graft-innervated gracilis functional free muscle transfer achieved a greater degree of emotionality during spontaneous smiling than masseteric or dually innervated transfer. Quantification of spontaneous smiling from standard video clips could facilitate future, blinded, multicenter trials with sufficient long-term follow-up to definitively establish the rates of spontaneity from a range of reanimation procedures.