Biofeedback System for Facial Palsy Follow-Up – INRLGII

Abstract

Introduction. Facial palsy is a neuromotor condition that alters facial mobility, symmetry, and expression, with functional, aesthetic, emotional, and social consequences. Its impact may be observed in daily activities such as speaking, smiling, closing the eyes, eating, or communicating through facial expressions. In clinical practice, facial palsy evaluation is commonly performed using observational scales and visual assessment by healthcare professionals. Although these tools are useful, they may present variability among evaluators and limit the objective quantification of changes during rehabilitation. Therefore, technological tools are needed to support therapeutic follow-up through quantifiable, reproducible, and accessible measurements. In this context, BIOPAF was developed as a biofeedback system for facial palsy assessment based on computer vision and artificial intelligence, integrating OpenCV, MediaPipe Face Mesh, and a digital clinical database to estimate facial asymmetry indexes.

Objective. The objective is to evaluate, through a proof-of-concept study, the preliminary ability of BIOPAF to analyze synthetic facial images generated by artificial intelligence and estimate facial asymmetry percentage. This approach made it possible to assess the initial technical performance of the system as a support tool for quantitative follow-up of facial alterations before its application to clinical images from real patients.

Methodology. A descriptive, cross-sectional proof-of-concept study was conducted using ten synthetic facial images generated by artificial intelligence. The images represented faces with different visual degrees of facial symmetry and asymmetry. BIOPAF was developed in Python 3.11, with a CustomTkinter graphical interface designed to facilitate its use by clinical personnel. Data management was implemented using SQLite3, allowing registration of records, analysis results, and asymmetry percentage. For facial analysis, the system integrated OpenCV and MediaPipe Face Mesh, selecting anatomical landmarks in the nose, mouth corners, and ocular region. The algorithm calculated Euclidean distances between central and bilateral points, comparing right and left facial regions to estimate facial asymmetry percentage. Results were organized according to the magnitude of the asymmetry percentage obtained for each image.

Results. BIOPAF processed the synthetic images and generated a facial asymmetry percentage for each case. Five images showed lower asymmetry values, ranging from 0.49% to 2.97%, while the remaining five showed higher values, ranging from 4.86% to 7.97%. The system enabled visualization of facial landmark detection, calculation of asymmetry indexes, and structured registration of automated analysis results. These findings show that BIOPAF can preliminarily differentiate AI-generated images with lower and higher facial asymmetry, maintaining quantitative separation between both image groups.

Conclusion. BIOPAF integrated computer vision, artificial intelligence, and digital records to objectively estimate facial asymmetry. The proof-of-concept study using synthetic images demonstrated the technical feasibility of the system as a support tool for quantitative facial palsy follow-up. However, since AI-generated images were used, the results should be interpreted as preliminary technical validation. A later stage using clinical images from real patients, comparison with established clinical scales, and definition of validated cut-off points is required to determine its usefulness in rehabilitation settings.