Accepted for/Published in: JMIR Mental Health
Date Submitted: Jan 26, 2026
Open Peer Review Period: Jan 26, 2026 - Mar 23, 2026
Date Accepted: Apr 4, 2026
(closed for review but you can still tweet)
Warning: This is an author submission that is not peer-reviewed or edited. Preprints - unless they show as "accepted" - should not be relied on to guide clinical practice or health-related behavior and should not be reported in news media as established information.
Validation of an Automated Video-Based Model for Detecting Tardive Dyskinesia Using the CTI Dataset
ABSTRACT
Background:
Tardive dyskinesia (TD) is a common, often underrecognized movement disorder resulting from long-term antipsychotic use, yet its detection in routine mental health care remains inconsistent despite the availability of structured rating scales.
Objective:
This study evaluates the performance of an AI-powered video-based model for detecting abnormal movements associated with tardive dyskinesia (TD) using the Clinician’s Tardive Inventory (CTI) dataset. We compare automated assessments of the videos in the CTI dataset with previously completed clinician-rated Abnormal Involuntary Movement Scale (AIMS) and CTI scores for the dataset’s videos to determine the model's reliability and the accuracy of its assessment conclusions relative to expert raters.
Methods:
Sixty-nine videos (n=69) with corresponding AIMS and CTI ratings were analyzed using the TDtect model previously reported in Sterns et al. The dataset included single-video assessments per participant, with varied instructions and movement types. The relationship between automated predictions and clinician ratings was assessed using Pearson's correlation, and predictive accuracy was evaluated using Area Under the Curve (AUC) metrics.
Results:
The model showed a strong correlation with AIMS total scores (r = 0.717) and high diagnostic accuracy (AUC = 0.854), which improved further at an optimized threshold (AUC = 0.900). Performance differed across anatomical regions, with the tongue, lips, and jaw displaying the highest predictive reliability. Functional CTI components had weaker correlations (r = 0.27-0.63), as expected due to the subjective nature of these measures.
Conclusions:
These findings validate the capability of AI-driven TD assessment using flexible clinical video protocols, broadening potential clinical applications. Further refinements and fine-tuning are expected to enhance accuracy, particularly in functional impact prediction. Clinical Trial: Remote Monitoring and Detecting of Tardive Dyskinesia for Improving Patient Outcomes, NCT06011408
Citation
Request queued. Please wait while the file is being generated. It may take some time.
Copyright
© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.