Machine Learning Methods to Track Dynamic Facial Function in Facial Palsy

Abstract

AbstractFor patients with facial paralysis, the wait for return of facial function and the resulting vision risk from poor eye closure, difficulty speaking and eating from flaccid oral sphincter muscles, as well as the psychological morbidity from the inability to smile or express emotions through facial movement can be devastating. There are limited methods to assess ongoing facial nerve regeneration: clinicians rely on subjective descriptions, imprecise scales, and static photographs to evaluate facial functional recovery and thus facial nerve regeneration remains poorly understood. We propose a more precise evaluation of dynamic facial function through video-based machine learning analysis which would facilitate a better understanding of the sometimes subtle onset of facial nerve recovery and improve guidance for facial reanimation surgery. Specifically, we here present machine learning methods employing likelihood ratio tests, optimal transport theory, and Mahalanobis distances to: 1) assess the use of defined facial landmarks for binary classification of different types of facial palsy; 2) identify regions of asymmetry and potential paralysis during specific facial cues; and 3) determining severity of abnormal facial function when compared to a reference class of normal facial function. Our work presents promising results of utilizing videos, rather than static photographs, to provide robust quantitative analyses of dynamic properties for various facial movements without requiring manual assessment. The long-term potential of this project is to enable clinicians to have more accurate and timely information to make decisions for facial reanimation surgery which will have drastic consequences on quality of life for affected patients.