From spasms to smiles: how facial recognition and tracking can quantify hemifacial spasm severity and predict treatment outcomes.

PURPOSE

Currently available grading and classification systems for hemifacial spasm either rely on subjective assessments or are excessively intricate. Here, we make use of facial recognition and facial tracking technologies towards accurately grouping patients according to severity and characteristics of the spasms.

METHODS

A retrospective review of our prospectively maintained preoperative videos database for hemifacial spasm was done. Videos were analyzed using an Apple AR kit-based App. A facial mesh is automatically allocated to specific biometric facial points. Videos are analyzed using Blender software for measuring the amplitude and frequency of the spasms. Classification of the patients into groups was done using both divisive k-means and agglomerative hierarchical clustering. Correlation-Analysis with preoperative quality of Life (Qol) using SF-36 questionnaire and HFS-8 score was performed. Additionally, correlation with postoperative outcome was calculated.

RESULTS

79 preoperative videos were included. Both up-bottom and bottom-up clustering approaches grouped the patients into 3 different clusters according to 4 variables (eye closure, mouth distance change, rate, and repetition of the spasms). Correlation of the groups with the Qol was done for 46/79 patients (58.2%). Spasms could be classified into mild, moderate clonic and severe tonic spasms. Patients with mild spasms showed better Qol scores. Moderate clonic spasms experienced best outcomes following microvascular decompression.

CONCLUSION

This novel classification using facial-tracking and augmented-reality is easy to use and apply. It quantifies the severity and type of the spasms and relates it to the quality of life of patients, postoperative outcome, and could guide our management strategy.