Automated Motor Tic Detection: A Machine Learning Approach.

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Video analysis aids objective assessment of tic frequency and severity in Tourette syndrome.
• Manual video ratings are time-consuming and impractical for large studies.
• Machine learning (ML) offers a solution for automated tic detection.

Purpose of the Study:

• Evaluate state-of-the-art ML approaches for automatic video-based tic detection.
• Assess the performance of ML algorithms in identifying tics in Tourette syndrome patients.

Main Methods:

• Two supervised ML approaches were developed: Random Forest with manual feature extraction and a deep learning convolutional neural network (CNN) with automated feature extraction.
• Utilized 64 videos from 35 Tourette syndrome patients, with 15 videos for validation.
• Binary classification task to distinguish between tic and no-tic segments.

Main Results:

• Random Forest achieved an F1 score of 82.0% (accuracy 88.4%).
• Deep neural network achieved an F1 score of 79.5% (accuracy 88.5%).

Conclusions:

• ML algorithms are feasible and reliable for automatic video-based tic detection.
• These tools can serve as valuable objective measures in clinical trials for Tourette syndrome.
• ML may also assist in the differential diagnosis of tic disorders.