AI and Tardive Dyskinesia: How Technology Is Advancing Care

Artificial intelligence is helping doctors spot subtle early signs of tardive dyskinesia, so fewer cases go undiagnosed and untreated.

Tardive dyskinesia affects hundreds of thousands of people, yet it often goes unrecognized. The symptoms can be easy to overlook at first — a subtle lip twitch, a slight hand movement, or a bit of restlessness that seems harmless. But for those taking antipsychotic medication, these small, involuntary movements may be the first signs of a serious neurological condition.

Diagnosing tardive dyskinesia isn’t easy. Symptoms can appear and disappear, vary from visit to visit, and be mistaken for something else entirely. Many cases go unnoticed until later in the disease course, when the movements have intensified and early treatment options can no longer slow progression. 

Now, developments in artificial intelligence (AI) may offer a way forward. Researchers are testing AI-powered tools — from video-based movement analysis to smartphone tracking apps — that could help identify the early, hardly noticeable signs of tardive dyskinesia. The hope is that technology can offer patients and doctors a clearer picture more quickly, leading to earlier care and better outcomes.

1. Video-Based AI Assessment

Modern AI systems can analyze videos of patients and detect movements that could signal tardive dyskinesia.

Scientists have recently created a machine learning algorithm that spotted tardive dyskinesia symptoms more accurately than trained humans. Publishing their findings in the Journal of Clinical Psychiatry, the researchers say AI could detect subtle symptoms that healthcare providers miss and reduce the impact of biases and fatigue that skew human judgment.

The team developed a vision transformer — a deep-learning model that classifies images and subtle movements. The model was trained on 3,979 videos from 351 people who completed the Abnormal Involuntary Movement Scale (AIMS) examination. AIMS includes tasks such as tapping your thumb with your fingers and extending your arms forward. 

Using a smartphone app, some participants self-recorded short videos in which they tapped a shoulder, opened their mouth and stuck out their tongue, sat still, and answered two questions.

When AI analyzed the videos, it was more accurate and consistent than human raters in determining who had tardive dyskinesia and who did not.

The researchers say their results show the potential of a video-based machine learning algorithm to monitor patients taking antipsychotic medications for signs of tardive dyskinesia. When a movement disorder is suspected, a trained care team can assess the symptoms and make a diagnosis, if necessary. 

2. Why Accurate Tracking Matters for Tardive Dyskinesia

When diagnosing tardive dyskinesia, healthcare providers check a patient’s medical history, including their use of dopamine receptor-blocking agents, according to a narrative review. Then, they conduct a neurological examination and a movement assessment. AIMS is often used to check for tardive dyskinesia symptoms and assess their severity.

The test is administered by a clinician trained to evaluate involuntary movements in body parts commonly affected by tardive dyskinesia, such as the face, extremities, and trunk. One drawback is that healthcare providers might rate the same patient differently — a challenge highlighted in a consensus statement. The reason? Different experts have varying levels of experience and might perceive the severity of movements differently. The authors of the statement suggested that researchers and clinicians may need to adopt new technologies to detect abnormal movements and other symptoms of tardive dyskinesia.

Another standard test is the Extrapyramidal Symptom Rating Scale (ESRS), which, like AIMS, relies on the clinician’s judgment and perception of symptom severity, according to a literature review. In the ESRS, individuals perform movements — such as walking 12 to 15 feet away and back or extending both arms forward with palms down and eyes closed — while the clinician observes and scores involuntary movements. Like AIMS, the ESRS depends heavily on the examiner’s clinical experience and interpretation.

A systematic review found that while these tools are recommended for diagnosing movement disorders, they have notable limitations. The researchers noted that AIMS is more effective at detecting facial movements than at detecting movements in other parts of the body. With the ESRS, their main concern was internal consistency — how reliably the scale items measure what they intend to.

If AI can outperform human raters, it may offer more objective and reproducible measurements that can help detect subtle patterns of abnormal movement. 

The scientists who developed the algorithm suggested that future studies could explore using AI for regular monitoring — such as monthly or quarterly assessments — so clinicians could detect early signs of tardive dyskinesia between visits and intervene sooner.

3. Remote and Home-Based Monitoring

In a recent review article, experts concluded that AI has the potential to be transformative in telemedicine by helping clinicians make more accurate diagnoses, continuously monitor patients remotely, and personalize care. For instance, AI has already been incorporated into telehealth for dermatology and ophthalmology to help find signs of skin cancer and diabetes-related eye disease.

When combined with in-person visits, telemedicine can improve both the diagnosis and treatment of tardive dyskinesia, according to an expert panel. Its benefits include reducing time and travel costs for patients, lowering no-show rates, and allowing caregivers and family members to participate in appointments. 

Because healthcare staff often face heavy workloads, conducting time-consuming movement checks for tardive dyskinesia can be challenging. As noted in the Journal of Clinical Psychiatry article, integrating AI into telemedicine could help detect the condition earlier and improve patient outcomes, particularly in areas with limited medical resources. 

4. How AI Could Transform Research and Treatment Discovery

AI’s ability to quickly process large, complex data sets — including detailed movement data — has made it an invaluable tool for researchers studying movement disorders. According to a systematic review, AI could help optimize clinical trials and support the development of new, more personalized treatments.

In clinical trials, the AIMS exam remains the gold standard for assessing tardive dyskinesia symptoms, with a 2-point change considered clinically meaningful, per a review. However, because it relies on subjective observation, results can vary. The Journal of Clinical Psychiatry article suggested that AI could provide more consistent, objective assessments.

AI systems can also capture fine details and timing that human raters might miss — analyzing each region of the body separately, identifying which areas move abnormally, and tracking when those movements occur. This type of continuous data collection could help researchers detect early signs of disease and inform clinical decisions.

For example, in one study, participants’ oral and facial movements were recorded both in a lab (using a 3D camera) and at home (using a tablet camera). A deep learning model then estimated movement features from the tablet videos and compared them with the lab data. The results showed that remotely collected data were consistent with in-lab findings and effectively tracked changes over time, suggesting AI could help clinicians monitor treatment progress or early signs of worsening symptoms.

5. AI and Personalized Medicine

AI could also help researchers fine-tune treatment for tardive dyskinesia through personalized medicine. In one study, scientists used AI to analyze AIMS scores from people taking either 40 or 80 milligrams (mg) of valbenazine (Ingrezza), a medication commonly prescribed to treat tardive dyskinesia. Using that data, their algorithm estimated how patients might respond to a middle-ground 60 mg dose. 

The analysis accurately predicted improvements in AIMS scores at that dosage — findings that helped support approval of the new 60 mg option by the U.S. Food and Drug Administration (FDA).

AI might even help identify who is most at risk of developing tardive dyskinesia while taking antipsychotic medication. In a study published in Schizophrenia Research, machine learning models helped scientists identify proteins and inflammatory markers detected in blood tests that correlated with a person’s odds of developing tardive dyskinesia. 

6. Creating Consistency in Tardive Dyskinesia Care

AI technology could help healthcare systems deliver more consistent, equitable care through precision medicine, a review concluded. For tardive dyskinesia specifically, AI may help detect the disorder more efficiently and accurately across large patient populations.

Experts on an FDA podcast also discussed how AI could be used to design efficient clinical trials, including decentralized trials. In these trials, people wouldn’t need to travel to a traditional study site; instead, data could be collected remotely, either from home or a nearby clinic or lab.
 
This approach could make research more accessible and inclusive by reaching communities that haven’t historically participated in clinical studies. It could also ensure that trial participants better represent the people who will ultimately benefit from new treatments.

7. Limitations of AI Diagnostics for Tardive Dyskinesia

While AI holds great promise for improving the diagnosis and treatment of tardive dyskinesia, it still has important limitations. 

• AI can generate inaccurate or biased information. AI models are only as reliable as the data they’re trained on. According to a review in the International Journal of Surgery, the accuracy and dependability of these tools depend on the quality and diversity of their training data. If that data is biased or incomplete, the AI’s output may also be biased, excluding underrepresented populations. For rare conditions like tardive dyskinesia, limited data can further reduce accuracy. In movement disorders, the complexity of abnormal movements can compound these issues, notes a review article published in Seminars in Neurology. Easy access to AI tools might also prompt some people to self-diagnose rather than seek professional care.
• AI cannot replace a doctor’s judgment. While AI can help clinicians spot early signs, it cannot diagnose tardive dyskinesia on its own. A trained clinician is still needed to interpret findings and consider important context, such as a person’s medical history and other relevant conditions.
• Video-based assessments may overlook certain body parts. Researchers writing in the Journal of Clinical Psychiatry found that AI could detect tardive dyskinesia from videos that excluded patients’ foot or toe movements. While rare, some people experience tardive dyskinesia only in the feet and toes — meaning these cases could be missed if a video-based exam doesn’t capture the full body.
• AI use raises privacy, data security, and consent concerns. Video monitoring can record a person’s face, voice, and even others nearby. Once collected, this sensitive data could be vulnerable to unauthorized access or misuse, such as through facial recognition technology or deepfakes, according to Seminars in Neurology. AI creators can reduce these risks by applying strong security measures, data encryption, and access controls, and by redacting or anonymizing video footage whenever possible.

Editorial Sources and Fact-Checking

• Deik A et al. The Role of AI in the Management of Movement Disorders. Seminars in Neurology. January 2025.
• Tardive Dyskinesia. Mental Health America. 
• Griffiths K et al. Identifying the Diagnostic Gap of Tardive Dyskinesia: An Analysis of Semi-Structured Electronic Health Record Data. BMC Psychiatry. April 21, 2025. 
• Tardive Dyskinesia. Cleveland Clinic. July 10, 2024.
• Correll C et al. Tardive Dyskinesia: Recognition, Patient Assessment, and Differential Diagnosis. Journal of Clinical Psychiatry. March 20, 2018.
• McEvoy J et al. Early Recognition and Treatment of Tardive Dyskinesia in Patients With Mood Disorders and Schizophrenia. Journal of Clinical Psychiatry. January 28, 2020. 
• Kremens D. Earlier Diagnosis of Tardive Dyskinesia. Journal of Clinical Psychiatry. December 10, 2019. 
• Sterns A et al. Detecting Tardive Dyskinesia Using Video-Based Artificial Intelligence. Journal of Clinical Psychiatry. May 28, 2025.
• Abnormal Involuntary Movement Scale. Oregon Health and Science University. 
• Faglioni A et al. Diagnosis and Management of Tardive Dyskinesia: From Research to Clinical Practice. Rivista di Psichiatria. March–April 2025.
• Nierenberg A et al. Revisiting the Abnormal Involuntary Movement Scale: Proceedings From the Tardive Dyskinesia Assessment Workshop. Journal of Clinical Psychiatry. May 8, 2018.
• Chouinard G et al. The Extrapyramidal Symptom Rating Scale and Its Abbreviated Version: A Critical Review of Clinimetric Properties. Psychotherapy and Psychosomatics. December 7, 2023.
• Martino D et al. Scales for Antipsychotic-Associated Movement Disorders: Systematic Review, Critique, and Recommendations. Movement Disorders. April 20, 2023. 
• Internal Consistency. APA Dictionary of Psychology. American Psychological Association.
• Rossi M et al. Integrating Artificial Intelligence Into Telemedicine: Evidence, Challenges, and Future Directions. Cureus. August 23, 2025.
• El-Mallakh R et al. Telehealth for Assessing and Managing Tardive Dyskinesia: Expert Insights From a Cross-Disciplinary Virtual Treatment Panel. Telemedicine and E-Health. July 4, 2023. 
• Vizcarra J et al. Artificial Intelligence in the Diagnosis and Quantitative Phenotyping of Hyperkinetic Movement Disorders: A Systematic Review. Journal of Clinical Medicine. November 21, 2024.
• Serrano D et al. Artificial Intelligence (AI) Applications in Drug Discovery and Drug Delivery: Revolutionizing Personalized Medicine. Pharmaceutics. October 14, 2024.
• Stacy M et al. Abnormal Involuntary Movement Scale in Tardive Dyskinesia: Minimal Clinically Important Difference. Movement Disorders. June 24, 2019. 
• Matthews D. From Assessment to Intervention: Evidence-Based Approaches in Tardive Dyskinesia. CNS Spectrums. February 12, 2025.
• Simmatis L et al. Reliability of Automatic Computer Vision-Based Assessment of Orofacial Kinematics for Telehealth Applications. Digital Biomarkers. July 21, 2022.
• Nguyen H et al. A Model-Informed Drug Development Approach Supporting the Approval of an Unstudied Valbenazine Dose for Patients With Tardive Dyskinesia The Journal of Clinical Pharmacology. July 25, 2024.
• Uludag K et al. Antipsychotic-Based Machine Learning Models May Help Prediction of Tardive Dyskinesia in Patients With Schizophrenia. Schizophrenia Research. February 2023.
• Bajwa J et al. Artificial Intelligence in Healthcare: Transforming the Practice of Medicine. Future Healthcare Journal. July 2021. 
• The Role of Artificial Intelligence in Clinical Trial Design and Research With Dr. ElZarrad. Q&A with FDA Podcast. Transcript. U.S. Food and Drug Administration. May 30, 2024.
• Chopra H et al. Revolutionizing Clinical Trials: The Role of AI in Accelerating Medical Breakthroughs. International Journal of Surgery. October 6, 2023.