Machine Learning Methods to Track Dynamic Facial Function in Facial Palsy.

Area of Science:

• Biomedical Engineering
• Computer Vision
• Neurology

Background:

• Facial palsy significantly impacts patients' quality of life, affecting vision, speech, eating, and emotional expression.
• Current methods for assessing facial nerve regeneration are subjective and imprecise, hindering timely clinical decisions.
• Objective and quantitative assessment tools are needed to monitor recovery and guide treatment.

Purpose of the Study:

• To develop and validate a machine learning-based system for precise, objective assessment of dynamic facial function in patients with facial palsy.
• To improve the understanding of subtle facial nerve regeneration onset.
• To enhance guidance for facial reanimation surgery.

Main Methods:

• Utilized machine learning techniques including likelihood ratio tests, optimal transport theory, and Mahalanobis distances.
• Employed defined facial landmarks for binary classification of facial palsy types.
• Identified asymmetric regions and quantified palsy severity, mapping it to clinical scores.

Main Results:

• Video analysis demonstrated significantly more accurate and detailed assessment of facial movements compared to existing methods.
• The developed methods allow for precise classification of facial palsy types.
• Objective quantification of palsy severity was achieved.

Conclusions:

• The proposed video-based machine learning analysis provides an objective and precise method for assessing facial palsy.
• This technology enables clinicians to make more informed and timely decisions regarding facial reanimation surgery.
• Improved assessment will lead to better patient outcomes and enhanced quality of life.