Artificial Intelligence in Facial Palsy Treatment: A Systematic Review and Recommendations.

Abstract

Background: Artificial intelligence (AI) is rapidly advancing and increasingly applied in facial palsy research. However, there is no comprehensive review to guide surgeons on AI-based facial assessment tools. While photographic standards exist, videographic standards for emotions have not been proposed. Implementing these standards is essential for improving information exchange and data comparison with the new AI tools.

Methods: We conducted a systematic review following the PRISMA guidelines, by analyzing databases including Medline, Embase, and the Cochrane Central Register of Controlled Trials. Our focus was on the use of AI-based facial assessment tools in patients with facial palsy who subsequently received intervention or surgery for their facial palsy. Data were evaluated descriptively, and recommendations, including videographic standards, were developed in collaboration with experts from multiple centers.

Results: We identified 3,222 articles, 35 of which met the inclusion criteria. Five AI applications analyzed static, dynamic, and chemodenervation procedures in unilateral or bilateral facial palsy. These focused on specific facial landmarks or emotion recognition from photos and videos, but varied in the expressions and emotions analyzed. Five studies provided validation data with either healthy subjects or other outcome measurements. We recommend a minimum videographic assessment including the validated emotions neutral and happy.

Conclusion: AI-related publications on facial palsy have significantly increased, but no consensus exists on the optimal AI-assessment software. The proposed flowchart from our systematic review can guide clinicians in decision-making. We recommend using the proposed videographic emotions to improve study consistency and comparability, while also encouraging further validation studies.