Zuzanna Miodońska

Background: Facial movement symmetry is an important indicator of neuromuscular function, with asymmetries associated with neurological disorders, trauma, and surgery. Quantitative symmetry assessment supports diagnosis, therapy monitoring, and surgical planning. This study proposes a marker-based approach to improve tracking stability and investigates whether dynamic facial movement descriptors can distinguish symmetric from asymmetric exercise execution. Methods: Videos were recorded using a low-cost acquisition setup during two facial exercises: eyebrow raising and smiling (75 patient; mean age 14 ± 4 years). Seventeen ArUco markers were placed at predefined facial landmarks. The dataset comprised 134 recordings labeled as symmetric (S) or asymmetric (AS). The processing pipeline included marker and face detection, symmetry axis estimation, feature extraction, and statistical analysis. Features were based on distances between paired markers and the estimated facial symmetry axis, yielding two dynamic descriptors: VertDist (vertical displacement) and Ratio (relative position across facial halves), along with their first derivatives. Results: Group differences between S and AS movements were analyzed using Welch’s t-test with effect sizes quantified by Hedges’ g. Statistically significant differences were found primarily in the first derivatives of VertDist and Ratio. For eyebrow raising, VertDist showed large effects (Hedges’ $\left|g\right|=1.41-1.42$) and Ratio moderate effects ($\left|g\right|=0.75-0.87$). For smiling, VertDist demonstrated moderate effects ($\left|g\right|=0.87-0.93$), while Ratio exhibited large effects ($\left|g\right|=1.14-1.21$). Conclusions: The proposed marker-based method enables reliable, low-cost quantitative assessment of facial movement asymmetry. Dynamic descriptors derived from VertDist and Ratio effectively differentiate symmetric and asymmetric facial movements.