A team of researchers at the University of California, San Francisco has developed a data-driven method for optimizing deep brain stimulation (DBS) settings that significantly improved walking performance in people with Parkinson’s disease. Published in the journal npj Parkinson’s Disease, the study used wearable sensors and implanted neural recording devices to analyze how different DBS settings affected walking, then applied machine learning to identify individualized stimulation parameters that enhanced gait. The results indicate that tailored DBS configurations can improve walking stability and speed, and suggest specific brain activity patterns linked to better mobility.
Parkinson’s disease is a progressive neurological condition that affects movement control. It arises from the loss of dopamine-producing neurons in the brain, leading to symptoms such as tremors, stiffness, slowness, and impaired balance. Gait disturbances—such as short shuffling steps, poor coordination, and freezing episodes—are among the most disabling symptoms, especially in later stages of the disease.
DBS is a surgical treatment in which electrodes are implanted in specific areas of the brain, typically the basal ganglia. These electrodes deliver electrical impulses to regulate abnormal brain activity. While DBS can be highly effective at reducing tremors and stiffness, its effects on walking tend to be inconsistent. This variability is partly due to the complexity of walking as a behavior, but also due to the lack of standardized methods for fine-tuning stimulation settings for gait.
DBS programming typically focuses on improving limb-related motor symptoms and is often conducted while the patient is seated. However, walking requires coordination across multiple brain regions, and existing programming practices do not consistently address gait. Complicating matters further, clinicians must choose from a wide range of stimulation parameters—amplitude, frequency, and pulse width—without clear guidance on how these settings affect walking behavior or neural dynamics.
The research team, led by Hamid Fekri Azgomi and Doris D. Wang, sought to overcome this challenge by designing a framework that integrates behavioral and neural data to guide DBS programming specifically for walking. Their goal was to uncover how different stimulation settings influenced both movement and underlying brain activity, and to create a predictive model that could identify optimal settings tailored to each individual.
“Gait disturbances are among the most disabling symptoms of Parkinson’s disease, severely affecting patients’ mobility, independence, and quality of life. While DBS has proven effective for alleviating other motor symptoms such as tremors and bradykinesia, its impact on gait remains unclear and inconsistent, making it challenging to determine optimal DBS settings for walking improvement,” said Wang, a functional neurosurgeon and an associate professor at UCSF and a faculty member in the UCSF–UC Berkeley Joint Graduate Program in Bioengineering.
“Recent advances in neurotechnology, including devices that can record brain activity while delivering stimulation in real time, have opened new opportunities to study the neural mechanisms underlying DBS. We launched this research to better understand how specific DBS parameters influence the brain circuits involved in walking. Our goal was to identify personalized stimulation settings that can help improve walking for individual patients with Parkinson’s disease.”
The study involved three people with Parkinson’s disease who had undergone DBS implantation targeting the globus pallidus (a part of the basal ganglia). Each participant also had electrodes placed over the motor cortex, a brain region involved in voluntary movement. These devices allowed simultaneous stimulation and recording of brain signals during walking tasks.
To capture the full range of motor behavior, participants were equipped with wearable sensors that tracked step length, stride speed, variability, and arm swing while walking in loops on a 6-meter track. The researchers developed a new measure called the Walking Performance Index (WPI), which combined these gait features into a single score that reflected overall walking ability.
The research team then systematically varied the DBS settings, adjusting amplitude, frequency, and pulse width within clinically safe ranges. For each configuration, participants completed walking trials while their brain activity and motion data were recorded. Subjective ratings from both patients and physical therapists were also collected.
A machine learning algorithm called a Gaussian Process Regressor was used to model the relationship between the stimulation parameters and WPI. This approach allowed the researchers to predict which combinations of settings would likely produce the best walking performance, even without testing every possible configuration. New predictions were tested in follow-up sessions and used to refine the model.
Each participant showed distinct patterns of gait improvement under different DBS configurations. The model successfully identified personalized settings that improved walking beyond what was achieved with standard clinical settings. One participant experienced an 18% improvement in walking performance with the new settings, while others showed smaller but meaningful gains.
“One surprising finding was the level of variability in how different DBS settings influenced gait. We expected some variation, but the extent to which each individual responded differently to specific stimulation patterns highlighted just how complex and patient-specific gait control is in Parkinson’s disease.”
In addition to behavioral data, the researchers analyzed neural signals from the brain during walking. They found that reductions in beta-band activity (a type of brain rhythm between 12–30 Hz) in the globus pallidus were consistently associated with better walking. These reductions were especially pronounced during specific phases of the gait cycle, such as when the opposite leg was bearing weight.
“The discovery that certain neural features were consistently associated with improved walking performance helped validate the potential of using brain signals to guide DBS therapy. These findings challenged the traditional one-size-fits-all approach and reinforced the importance of letting the brain guide therapy, highlighting the need for personalized, adaptive stimulation strategies informed by each patient’s neural activity.”
Importantly, while the precise patterns of improvement varied across individuals, the machine learning model was able to adapt to each person’s unique neural and motor responses. In follow-up periods, one participant voluntarily used the model-recommended settings for several hours each day, demonstrating the practicality of implementing these changes outside the lab.
The study highlights the importance of personalized DBS programming for addressing gait disturbances in Parkinson’s disease. Standard programming practices may not account for the complex and individual-specific brain dynamics involved in walking. By using real-time brain recordings and data-driven modeling, this approach offers a path toward more effective and targeted treatments.
“The most important takeaway is that brain stimulation for Parkinson’s disease can, and should, be personalized. Our study shows that gait improvement is not just about turning DBS on or off, but about finding the right settings designed to everyone’s brain activity and walking patterns.”
“By modeling the relationship between stimulation parameters, neural activity, and gait performance, we demonstrate the potential for data-driven, individualized DBS therapies that go beyond standard approaches. This opens the door to more precise and effective treatment strategies, particularly for challenging symptoms like gait dysfunction that have not responded consistently to conventional DBS. This work lays the foundation for adaptive DBS systems that adjust therapy in real time, based on how the patient’s brain and body respond, bringing us one step closer to intelligent neuromodulation in everyday care.”
The study was limited by its small sample size—only three participants were included. While the results are promising, larger studies are needed to confirm the findings and assess how well this approach generalizes to more diverse patient populations. In addition, the study focused on straight walking; future work should explore how turning, freezing, and obstacle navigation respond to tailored stimulation.
“While our results are promising and highlight the potential of personalized DBS to improve gait in Parkinson’s disease, it is important to recognize that this study was conducted in a small cohort of patients. These findings do not imply that a single DBS setting will universally restore gait. Instead, the takeaway is that data-driven, individualized approaches, grounded in both neural signals and behavioral metrics, can offer a more systematic and responsive way to optimize therapy.”
The researchers hope to expand their work by incorporating longer walking trials, larger datasets, and more advanced algorithms. They envision future DBS systems that use neural biomarkers and machine learning to continuously adapt stimulation in real-time, improving mobility throughout the day. This could dramatically reduce the burden on patients and clinicians during programming sessions and improve quality of life for people with Parkinson’s disease.
“We hope this work lays the foundation for intelligent, adaptive DBS systems that continuously adjust therapy based on a patient’s real-time brain and movement signals. The identified gait-optimized DBS settings have the potential to inform future adaptive DBS designs and move clinical practice beyond static programming toward closed-loop systems that respond to each individual’s dynamic needs throughout the day. Additionally, the neural features associated with improved walking performance could support clinicians during programming visits, making the process more efficient and objective. Similar data-driven approaches could be extended to optimize treatment for a broader range of motor and non-motor symptoms.”
“This study represents a collaborative effort between clinicians, engineers, and neuroscientists, highlighting the value of interdisciplinary work in advancing personalized neuromodulation therapies. Our findings offer a step toward more adaptive DBS systems for gait dysfunction and illustrate the power of integrating neural signals, behavioral metrics, and machine learning to tailor treatments to individual needs.”
The study, “Modeling and optimizing deep brain stimulation to enhance gait in Parkinson’s disease: personalized treatment with neurophysiological insights,” was authored by Hamid Fekri Azgomi, Kenneth H. Louie, Jessica E. Bath, Kara N. Presbrey, Jannine P. Balakid, Jacob H. Marks, Thomas A. Wozny, Nicholas B. Galifianakis, Marta San Luciano, Simon Little, Philip A. Starr, and Doris D. Wang.
