AI Model Advances Automated Video-Based Assessment of Facial Palsy
May 29, 2025
A refined AI tool using facial landmark detection improves the objective evaluation of facial palsy severity in clinical videos.
Key Details
- Researchers fine-tuned a facial recognition AI model (3D-FAN) for patients with facial palsy using 1,181 images from 196 patients.
- Manual annotation of facial keypoints improved the model's accuracy, particularly for eyelids and mouth asymmetry.
- The refined tool showed lower error rates in keypoint detection compared to baseline models trained on healthy faces.
- Objective ratings from the model may aid treatment planning and outcome assessments.
- Authors plan to make the AI model freely available for wider clinical and research use.
Why It Matters
Objective and automated evaluation tools for facial palsy can standardize assessments and improve outcomes, demonstrating a valuable extension of imaging AI in clinical neurology and surgery. Sharing these tools openly can accelerate research and adoption for other rare disorders as well.