Kristina Simonyan, MD, PhD, DrMed of Mass Eye and Ear and the Mass General Brigham Neuroscience Institute, is the senior author of a paper published in npj Digital Medicine, " DystoniaDBSNet as a novel deep learning biomarker of predictive deep brain stimulation outcome in dystonia ."
Q: What challenges or unmet needs make this study important?
Less than 5% of patients with dystonia — a debilitating neurological movement disorder — undergo deep brain stimulation (DBS) treatment, even though it is one of the few interventions that directly target the abnormal brain processes underlying the condition. The surgical treatment is highly underutilized, likely due to its invasiveness and the lack of objective markers of who may benefit. Even among those who undergo surgery, outcomes vary widely, depending on patient characteristics and external influences.
Q: What central question were you investigating?
Currently, no neural biomarkers exist that help clinicians predict whether a patient with dystonia will benefit from DBS. Therefore, we set out to develop an automated platform that could objectively predict DBS treatment response by identifying relevant patterns in a patient's brain MRI.
Q: What methods or approach did you use?
We developed DystoniaDBSNet, a deep learning algorithm that analyzes patients' pre-operative MRIs to identify brain patterns associated with treatment response and predicts their likelihood of benefiting from DBS. Importantly, it makes this prediction before surgery, when it can help inform candidate selection and treatment planning.
The algorithm was trained, independently tested and externally validated in an international cohort from leading DBS Centers. Its development builds on years of our research aimed at understanding the brain mechanisms underlying dystonia.
Q: What did you find?
The algorithm identified a distinct brain pattern associated with DBS treatment effectiveness and used this pattern to predict how patients would respond to surgery. DystoniaDBSNet achieved 95.5% accuracy and showed strong predictive performance across different forms of dystonia, generating a prediction in under a minute per patient.
Q: What are the real-world implications, particularly for patients?
Because it is fast, easy to use and relies on routinely acquired clinical brain MRIs, the algorithm has strong potential to help clinicians identify patients most likely to benefit from DBS and guide treatment planning. More broadly, this approach could expand access to DBS for patients with different forms of dystonia.
Follow-up studies will focus on translating the DystoniaDBSNet platform into clinical practice and further evaluating its performance in real-world settings. The ultimate goal is to help patients regain control over their everyday movements.
Q: What was your most memorable moment during the study?
One particularly memorable case involved a patient for whom DystoniaDBSNet predicted a 99% probability of benefiting from DBS, yet the patient initially showed no clinical improvement. At first, we assumed the algorithm had misclassified this patient. However, following additional surgical intervention, the patient experienced significant symptom improvement consistent with the algorithm's original prediction.
This case was especially striking because it illustrated the potential of DystoniaDBSNet to complement clinical evaluation and help inform treatment decisions even when a patient's initial outcome is unexpected.
Q: When you aren't in the lab, how do you enjoy spending time?
Classical music is a big part of our family, with my kids studying violin, viola and piano. I rarely miss their lessons and recitals; it gives me great joy to watch them grow as musicians. I also love spending time in my garden, where I take care of our trees and flowers and grow nearly our entire summer crop of vegetables and berries.
Authorship: In addition to Simonyan, Mass General Brigham authors include Dongren Yao, Giovanni Battistella and Nutan Sharma. Additional authors include Emad Eskandar, Eoin Mulroy, Kailash Bhatia, Joshua Wong, Jun Yu, Andres Deik, Thomas Foltynie, Patricia Limousin, Harith Akram, Ludvic Zrinzo, Gordon Baltuch, Meredith Spindler, Kilsoo Kim and Shiro Horisawa.
Disclosures: Simonyan reports receiving funding from the National Institutes of Health, the Department of Defense and Amazon Web Services; serving on the Scientific Advisory Boards of the Tourette Association of America and The Voice Foundation; and being an inventor on a granted U.S. patent and pending European patent relevant to this work, as well as pending U.S. patents unrelated to this work. Additional author disclosures can be found in the paper.
Funding: This study was funded by the National Institute of Neurological Disorders and Stroke and the National Institute on Deafness and Other Communication Disorders, the National Institutes of Health (R01NS124228, R01DC011805, P50DC019900) and the Amazon Web Services Machine Learning Research Award.
Paper Cited: Yao, D., et al. "DystoniaDBSNet as a novel deep learning biomarker of predictive deep brain stimulation outcome in dystonia." npj Digital Medicine. DOI: 10.1038/s41746-026-03235-8