Deep learning-based automatic facial symmetry scoring in peripheral facial palsy

Area of Science:

• Medical Imaging
• Computer Vision
• Rehabilitation Science

Background:

• Unilateral peripheral facial palsy (PFP) causes facial asymmetry and functional deficits, impacting patient quality of life.
• Objective assessment tools are crucial for effective PFP monitoring and rehabilitation strategies.
• Current assessment methods may lack objectivity and precision in quantifying facial movement symmetry.

Purpose of the Study:

• To develop and validate an automated method for objective facial symmetry assessment in PFP patients.
• To quantify facial movement symmetry using heatmaps and symmetry scores derived from standardized 2D photographs.
• To correlate automated symmetry scores with clinical Stennert movement scores.

Main Methods:

• Utilized a dataset of 405 facial images from 198 PFP patients.
• Employed deep learning for facial landmark detection and an affine alignment algorithm.
• Generated heatmaps from grayscale difference images and calculated symmetry scores by comparing facial halves.

Main Results:

• The automated method successfully processed all datasets, yielding symmetry scores from 0 to 0.99 (mean 0.85 ± 0.12).
• Heatmaps visually represented asymmetries consistent with clinical observations.
• Significant negative correlations (r = -0.32 to -0.66) demonstrated that higher clinical severity correlated with lower symmetry scores.

Conclusions:

• The developed automated method offers an objective, reliable, and accessible tool for assessing facial symmetry in PFP.
• This approach enhances clinical evaluation accuracy and enables precise monitoring of rehabilitation progress.
• The method shows potential for higher sensitivity in detecting subtle changes compared to traditional scoring.