Automated Quantification of Eye Tics Using Computer Vision and Deep Learning Techniques.

Area of Science:

• Computational neuroscience
• Medical imaging analysis
• Machine learning in healthcare

Background:

• Traditional Tourette syndrome (TS) tic quantification relies on subjective rating scales.
• Existing objective video-based methods are resource-intensive and require human raters.
• Computer vision offers automated detection of atypical movements for tic quantification.

Purpose of the Study:

• To apply a computer vision approach to train a supervised deep learning algorithm.
• To detect eye tics, the most common tic type in TS patients, from video data.

Main Methods:

• Utilized 54 videos from 11 adolescent TS patients.
• Human raters identified 1775 eye tic events and 3680 non-tic events.
• Applied supervised deep learning to 3D facial landmarks from video clips.

Main Results:

• Achieved an Area Under the Curve (AUC) of 0.89 for eye tic classification using a random split regimen.
• Demonstrated an AUC of 0.74 for a disjoint split regimen, indicating limited generalizability with small patient samples.
• The algorithm successfully detected eye tics in unseen validation data.

Conclusions:

• Automated eye tic detection via computer vision is feasible and accurate.
• This technology shows potential for improving tic quantification in TS.
• Future applications include TS screening, diagnostics, and treatment outcome assessment.