Innovative AI Technology Enhances Automated Assessment of Facial Palsy

A new, finely-tuned artificial intelligence (AI) system shows significant potential for providing objective evaluations of facial palsy severity, according to a recent study published in the June edition of Plastic and Reconstructive Surgery. Patients suffering from facial palsy experience paralysis or weakened movement on one or both sides of the face, often resulting from nerve damage caused by tumors, surgical procedures, or trauma. Accurate assessment of the condition is crucial for planning effective treatments, such as nerve transfer surgeries, but has historically posed challenges due to subjective scoring systems that vary between clinicians.

Traditional objective assessment methods are often impractical for everyday clinical use, prompting researchers to explore machine learning and AI as alternatives for routine, reliable evaluation. The study focused on improving an existing facial recognition AI model, known as 3D-FAN, which was initially trained to identify 68 facial keypoints, including eyebrows, eyelids, nose, mouth, and facial contours. When tested on clinical videos, however, the model struggled to accurately detect facial asymmetries characteristic of palsy, especially in cases involving facial expressions like smiling or eyelid closing.

To address this, the research team, led by Dr. Takeichiro Kimura from Kyorin University, applied machine learning techniques to refine the AI. They manually adjusted facial landmarks in 1,181 images from 196 patients' videos, minimizing variability and retraining the model until improvements plateaued. This process demonstrated notable enhancements in the AI's ability to detect facial keypoints, including in critical areas affected by palsy.

The refined model exhibited markedly lower error rates across all facial regions, especially around the eyelids and mouth. These improvements suggest that the AI can now provide more precise, objective severity scores for facial palsy, which can be invaluable for clinical decision-making and tracking treatment outcomes. Dr. Kimura emphasized that this approach—combining manual landmark correction with machine learning—may extend to other rare disorders requiring specialized assessments.

Looking ahead, the team plans to make their AI system freely accessible to clinicians and researchers, further exploring its potential to serve as a reliable tool for facial palsy evaluation. By offering an objective and quantitative assessment method, this technology could significantly enhance the accuracy of diagnoses, treatment planning, and monitoring of facial paralysis, ultimately improving patient care.

Source: Medical Xpress