Implanted Brain Devices Detect Walking at Home in Parkinson's Patients for the First Time

Deep brain stimulation has become a standard treatment for Parkinson's disease, but it operates as a blunt instrument. Current devices deliver continuous electrical pulses to movement-related brain structures at fixed settings - the same stimulation whether the patient is walking, sleeping, eating, or sitting still. That approach works reasonably well for tremor and stiffness, but gait impairment, one of the most disabling aspects of Parkinson's disease, often responds poorly to DBS settings optimized for other symptoms.

A smarter device would do what a skilled clinician does: adjust based on what the patient is actually doing at any given moment. To get there, researchers first need to answer a more basic question - can an implanted device read brain signals accurately enough to determine a patient's movement state during ordinary daily life, without the controlled conditions of a laboratory? A study published February 13 in Science Advances by a UC San Francisco team demonstrates, for the first time, that the answer is yes.

Four Patients, 80 Hours, Unsupervised

The study enrolled four participants with Parkinson's disease who had been implanted with an investigational bidirectional DBS system - a device capable of both delivering stimulation and recording neural activity from brain electrodes. That recording capability is not available in standard clinical DBS devices currently approved for patient use.

The implanted electrodes captured signals from two movement-related brain regions: the motor cortex, which plans and initiates movement, and the globus pallidus, a deep structure involved in movement regulation that is a standard DBS target for Parkinson's. Simultaneously, wearable accelerometers on the patients' limbs provided ground-truth movement data - identifying when participants were actually walking versus sitting, standing, or engaged in other activities.

Altogether, the team collected and analyzed synchronized neural and movement data spanning more than 80 hours of unsupervised activity across the four participants. This was not a laboratory protocol: patients went about their normal daily lives at home, providing a real-world signal environment with all the variability - ambient noise, casual movements, changing emotional states - that clinical conditions carefully exclude.

Neural Signatures Are Individual, but Reliable

The analysis found that brain activity patterns associated with walking were detectable from the implanted electrodes - but that those patterns varied across individuals. Each patient had a distinct neural signature for walking, meaning that algorithms trained on one patient's brain data would not necessarily generalize to another's. The team therefore developed individualized classifiers for each participant rather than a single universal model.

Within each individual, the neural walking signatures were consistent enough to allow the implanted device to classify movement states - walking versus not walking - in real time using the brain signal alone, without relying on the wearable sensors that provided the ground-truth data during this phase of the research.

"We identified personalized neural biomarkers associated with gait and demonstrated that these signals can be used for real-time movement state classification within the constraints of an implanted device," said senior author Doris Wang, MD, PhD, a neurosurgeon and associate professor of Neurological Surgery at UCSF. "This establishes a framework for future adaptive DBS systems that could adjust stimulation in response to a patient's activity state."

Why Gait Is a Different Problem

Gait symptoms in Parkinson's disease include shuffling steps, freezing of gait (sudden inability to initiate movement), instability during turning, and difficulty on uneven terrain. These symptoms fluctuate throughout the day and are often worst when dopamine levels are low - for example, in the hours before medication doses take effect. They are also unpredictable: a patient may walk well for most of a morning and then freeze while approaching a doorway.

Standard continuous DBS does not address that fluctuation because it cannot sense the patient's state. An adaptive system that detects the onset of walking - or the neural precursors to gait initiation - could in principle apply optimized stimulation parameters specifically for walking, then switch to different settings appropriate for rest, sleep, or fine motor tasks. The UCSF study provides the neural detection component that such a system would require.

Early Stage, Small Sample

The authors are explicit that this was a feasibility study, not a clinical efficacy trial. Four participants is a very small sample for drawing generalizable conclusions about which neural features best predict gait, how reliably classification works across the full range of Parkinson's disease presentations, or how well individualized classifiers would perform over months and years as the disease progresses and neural signals change.

The study also did not test whether actually adjusting stimulation based on the detected walking state improved patient outcomes. That critical next step - demonstrating that closed-loop, activity-responsive DBS actually helps patients walk better, fall less, or live more independently - requires larger trials with clinical endpoints.

The research was supported by the Michael J. Fox Foundation, NIH grant R01NS130183, a UCSF Catalyst Grant, and the Tianqiao and Chrissy Chen Institute. Planning for follow-up trials testing adaptive stimulation is underway at UCSF.