Related Experiment Video
Updated: Dec 28, 2025
05:21
Computerized Adaptive Testing System of Functional Assessment of Stroke
Published on: January 7, 2019
6.2K
Toward an Automatic System for Computer-Aided Assessment in Facial Palsy
Diego L Guarin1,2, Yana Yunusova1,3,4, Babak Taati1,5,6
1KITE | Toronto Rehabilitation Institute-University Health Network, Toronto, Canada.
Facial Plastic Surgery & Aesthetic Medicine
|February 14, 2020
Summary
Machine learning models for facial landmark detection show improved accuracy for facial palsy patients when retrained with fewer than 1600 patient images. This advancement aids in automated computer-aided diagnosis for facial palsy.
Area of Science:
- Computer Vision
- Medical Imaging
- Machine Learning
Background:
- Quantitative facial function assessment is limited by subjective grading scales.
- Machine learning (ML) offers automated quantification of facial metrics from images.
- Current ML models for facial landmark localization need improvement for clinical application.
Purpose of the Study:
- Develop a novel ML algorithm for rapid and accurate facial landmark localization in facial palsy patients.
- Integrate this technology into an automated computer-aided diagnosis system.
Main Methods:
- Manually annotated 68 facial landmarks in 8 expressions from 200 facial palsy patients and 10 healthy controls.
- Trained a novel ML model using a disease-specific database of patient photographs.
- Compared algorithm accuracy against manual markings and a model trained on healthy subjects.
Main Results:
- Publicly available algorithms showed poor accuracy on patient images (NRMSE 8.56 ± 2.16) compared to healthy controls (7.09 ± 2.34).
- A model retrained with 1440 patient images significantly improved localization accuracy (NRMSE 6.03 ± 2.43) versus a model trained on thousands of healthy faces (8.56 ± 2.16).
Conclusions:
- Retraining computer vision models with <1600 annotated patient images substantially enhances facial landmark detection performance.
- The developed database and ML model are foundational for automated facial palsy assessment systems.
