Using AI to Measure Eye and Face Features
Author Information
Author(s): Rana Khizar, Beecher Mark, Caltabiano Carmelo, Macri Carmelo, Zhao Yang, Verjans Johan, Selva Dinesh
Primary Institution: University of Adelaide
Hypothesis
Can a deep learning model accurately automate the assessment of periocular measurements?
Conclusion
The automated facial landmark detection network provided accurate and reliable periocular measurements.
Supporting Evidence
- The AI algorithm demonstrated close agreement with human measurements, with mean absolute errors ranging from 0.22 mm to 0.88 mm.
- Intraclass correlation coefficients indicated excellent reliability for most measurements.
- The landmark detection model achieved a mean error rate of 0.51%.
Takeaway
This study created a smart computer program that can measure parts of the face around the eyes very accurately, which can help doctors do their jobs better, especially when seeing patients online.
Methodology
Participants had their images taken, and facial landmarks were segmented to calculate various periocular measurements, which were then used to train a machine learning algorithm.
Potential Biases
The study may have biases related to the specific patient population and the controlled imaging conditions.
Limitations
The study was conducted at a single center, which may limit generalizability, and the AI system needs external validation on images from different cameras and settings.
Participant Demographics
The mean age of participants was 59 years, with 54% female and a majority being Caucasian.
Statistical Information
P-Value
<0.0001
Confidence Interval
95% CI for ICCs ranged from 0.883 to 0.998
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website