Researchers of this multi-phase study evaluated a video-based artificial intelligence (AI) algorithm designed to detect and assess the severity of tardive dyskinesia (TD) in patients taking antipsychotics. Across three studies involving 351 participants and nearly 4,000 video clips, the algorithm analyzed facial and upper body movements captured via smartphone. Compared to clinician-rated Abnormal Involuntary Movement Scale scores, the model achieved a high area under the curve of 0.85 to 0.98 and outperformed trained raters in sensitivity, specificity, and reliability (Cohen’s κ = 0.61 vs 0.57). It also demonstrated consistent performance across demographic subgroups and could pinpoint specific body regions with abnormal movements.

The findings highlight the algorithm’s potential to close the diagnostic gap for TD, which affects up to 2.6 million people in the United States, though only about 40,000 currently receive treatment. With psychiatric care strained by limited time and resources, this scalable AI tool offers an efficient solution for earlier detection and monitoring. Early identification enables timely interventions, whether this be medication adjustments or symptom management, before TD becomes irreversible. As such, this technology could improve outcomes and reduce the burden of long-term disability in patients at risk for TD.

Reference: Sterns AA, Hughes JW, Grimm B, et al. Detecting Tardive Dyskinesia Using Video-Based Artificial Intelligence. J Clin Psychiatry. 2025;86(3):25m15792. doi: 10.4088/JCP.25m15792.