Continuous and componentized facial palsy measurement alignment and clinical interpretable model

Area of Science:

• Medical Imaging and Computer Vision
• Neurology
• Ophthalmology

Background:

• Facial palsy affects 1 in 60 individuals, necessitating accurate assessment for effective treatment.
• Current facial palsy grading systems lack consistency due to subjective factors and fail to address separate nerve innervations for eyelids and mouth.

Purpose of the Study:

• To develop and validate a modified House-Brackmann (H-B) criteria system using continuous and componentized measurements for eyelid and mouth movements.
• To create a precise and comprehensive assessment tool for facial palsy.

Main Methods:

• A dense facial landmark alignment model was developed using a dataset of 274 patients with vestibular schwannoma (VS).
• The model generated clinically interpretable asymmetric coefficients for eyelids and mouth.
• Receiver Operating Characteristic (ROC) analysis thresholds transformed coefficients into grades comparable to consensus H-B grades.

Main Results:

• The developed model outperformed existing algorithms in facial palsy detection.
• Eyelid and mouth coefficients showed strong correlation with consensus H-B grades (r=0.892 and 0.890, P<0.001).
• Validation in a separate multi-center cohort demonstrated high accuracy.

Conclusions:

• The proposed model offers a more precise and comprehensive assessment of facial palsy.
• It has the potential for continuous monitoring of facial palsy progression and treatment response.
• This approach addresses limitations of current subjective grading systems.