Automated Video-Based Approach for the Diagnosis of Tourette Syndrome

Background: The occurrence of tics is the main basis for the diagnosis of Gilles de la Tourette syndrome (GTS). Video-based tic assessments are time consuming. Objective: The aim was to assess the potential of automated video-based tic detection for discriminating between videos of adults with GTS and healthy control (HC) participants. Methods: The quantity and temporal structure of automatically detected tics/extra movements in videos from adults with GTS (107 videos from 42 participants) and matched HCs were used to classify videos using cross-validated logistic regression. Results: Videos were classified with high accuracy both from the quantity of tics (balanced accuracy of 87.9%) and the number of tic clusters (90.2%). Logistic regression prediction probability provides a graded measure of diagnostic confidence. Expert review of about 25% of lower-confidence predictions could ensure an overall classification accuracy above 95%. Conclusions: Automated video-based methods have a great potential to support quantitative assessment and clinical decision-making in tic disorders.