Facial paralysis is a loss of facial movement due to nerve damage. It is essential for clinicians to diagnose the severity of the facial paralysis to treat patients, assess progresses and evaluate outcomes. Subjective assessments are common in clinical practices but have their limitations regarding the intra-observer and inter-observer reproducibility. We utilised the dynamic 3D stereo photogrammetry technology for the objective grading of facial paralysis by measuring regional facial asymmetries. The correlations between the measured asymmetries and scores of a modified Sunnybrook facial paralysis grading were evaluated to identify the region of interests of objective measurements closely related to the subjective grades. Categorical classifiers were trained to quantify the severity of the facial paralysis. Preliminary results showed that the objective asymmetry measurements were highly correlated to the subjective assessments of facial paralysis except the eye region. Machine learning approaches showed a potential of improving the accuracy of severity assessments.

面瘫是由于神经损伤导致面部运动丧失。临床医生必须诊断面瘫的严重程度以治疗患者、评估进展和评估结果。主观评估在临床实践中很常见，但在观察者内和观察者间的可重复性方面存在局限性。我们利用动态 3D 立体摄影测量技术通过测量局部面部不对称来对面部麻痹进行客观分级。评估测量的不对称性与改良 Sunnybrook 面瘫分级分数之间的相关性，以识别与主观评分密切相关的客观测量的兴趣区域。训练分类分类器来量化面瘫的严重程度。初步结果表明，客观不对称测量与除眼部区域外的面瘫主观评估高度相关。机器学习方法显示出提高严重性评估准确性的潜力。