The Era of Learning to Walk with Deep Brain Stimulation ─ AI Treatment Improves Gait in Parkinson's Disease Patients

In the treatment of Parkinson's disease, there has long been a "remaining challenge." Medications and deep brain stimulation can sometimes be effective for tremors, muscle stiffness, and slowness of movement. However, gait disturbances, particularly "freezing of gait" where the feet feel glued to the ground, as well as instability in daily movements like turning, standing up, and climbing stairs, have significantly limited patients' lives.

This time, the German newspaper WELT covered a new AI-assisted deep brain stimulation therapy developed by a Swiss research team. The research was led by scientists from the Swiss Federal Institute of Technology Lausanne and Lausanne University Hospital. The published paper demonstrated a system that reads the brain activity of Parkinson's patients in real-time, estimating whether the person is sitting, standing, walking, or trying to avoid an obstacle, and automatically adjusts the electrical stimulation to the brain accordingly.

In other words, while previous deep brain stimulation was a "treatment that continuously sends stimulation at a constant rhythm," this new technology is a "treatment that changes stimulation according to the patient's movements." The research team positions this as a blueprint for next-generation neuromodulation, a treatment that adjusts neural activity according to the situation.

In Parkinson's disease, degeneration of nerve cells in the brain leads to a deficiency of dopamine, disrupting the mechanism that smoothly controls movement. The typical symptoms are tremors, muscle stiffness, and slowness of movement, but as the disease progresses, gait and posture problems become severe. Particularly, freezing of gait causes the feet to feel glued to the floor regardless of the person's will, leading to anxiety about falling or going out. A few steps inside the house, a step at the entrance, a crowded station, turning in a narrow passage—what seems like a trivial movement to healthy people becomes a significant barrier for patients.

Deep brain stimulation is a treatment that involves inserting electrodes into specific parts of the brain and sending electrical signals from a stimulator implanted in the chest or other areas. It has decades of clinical experience and has been used for the motor symptoms of advanced Parkinson's disease. It is often effective for tremors, muscle rigidity, and slowness of movement. However, its effectiveness for gait disturbances has been unstable, and in some cases, it has even been reported to worsen the condition.

One reason for this is that walking is not a simple movement. Sitting, standing, starting to walk, turning, avoiding obstacles, climbing stairs—each requires different neural control. Moreover, in Parkinson's disease, the effectiveness of medication fluctuates depending on the time of day, and the physical condition of the same patient can change significantly from morning to evening, before and after taking medication. It is challenging to address all these variations with fixed stimulation parameters.

The key point of this research is the utilization of clues to decipher walking states from signals recorded from a brain region called the subthalamic nucleus. The research team simultaneously recorded brain activity, whole-body movement, and leg muscle activity in 35 patients with advanced Parkinson's disease. Patients were asked to perform tasks such as sitting, standing, walking, and avoiding obstacles, and the changes in brain signals during these tasks were analyzed.

It was found that brain activity exhibits characteristic patterns depending on the type of movement. Furthermore, using machine learning, it became possible to estimate the patient's motor state in real-time from brain signals. Importantly, this estimation was not just a classification experiment in the laboratory but was actually applied to stimulus control.

The research team tested activity-dependent adaptive deep brain stimulation on four patients with advanced Parkinson's disease. All subjects were patients who still had gait disturbances despite receiving conventional optimized treatment. The new system analyzes the characteristics of brain signals for each patient and creates rules for adjusting stimulation that suit them. Some patients showed improved leg movement when the stimulation was increased during walking, while for others, reducing the stimulation was more beneficial. In other words, the AI is not making a simple judgment of "stronger stimulation is better." It is a personalized treatment that changes the balance of stimulation according to the patient's condition and movements.

In the trial, improvements were observed in gait stability, stride length, turning, frequency of freezing of gait, ease of standing up, and continuation of walking. One participant, who previously felt their legs were heavy and sometimes uncontrollable, reported being able to walk longer as the stimulation changed according to their movements. This is not just about improved test scores. For patients, it translates to "being able to walk a little further," "not being afraid to stand up," and "reduced anxiety about going out," directly impacting their daily freedom.

The attention this news has garnered is partly due to the impression of the term "AI healthcare." Many people associate AI with image diagnosis, chatbots, and drug discovery support. However, in this technology, AI does not provide advice from behind a screen but interacts with a device implanted in the patient's body, reading brain activity and changing stimulation. In other words, AI is directly involved in the timing and intensity of treatment.

However, it is important not to misunderstand that AI is independently deciding the treatment plan. The system operates within a framework designed by doctors and researchers, adjusting stimulation based on verified neural signals for each patient. In healthcare, AI functions more as an assistant that captures millisecond-level changes that human eyes and hands cannot track, allowing for fine-tuned treatment adjustments, rather than replacing doctors.

On social media, there are voices of expectation for this research. The LinkedIn post by the researcher himself received multiple comments and was shared mainly among experts in neuroengineering and deep brain stimulation. Reactions were also gathered on the official posts by EPFL and CHUV, with notable perceptions like "it addresses gait disturbances that traditional DBS struggled with" and "real-time adaptive treatment is closer to daily life." On X, the study was shared through the Nature Medicine article introduction and DBS-related hashtags, with users in the neurotechnology field expressing reactions like "exciting research."

On the other hand, the reactions on social media are not all unreserved praise. Medical and research-oriented users, while evaluating the achievement as a "significant step," also take a cautious view, noting that "it is still a small-scale demonstration," "long-term safety and efficacy need to be confirmed," and "stability in real-life environments is crucial." This is a very healthy reaction because the clinical trial was a feasibility test involving four people, and it is not yet at the stage where it can be used as a general treatment.

There are still several challenges to be addressed for practical application. First, the effectiveness needs to be confirmed in more patients. The symptoms of Parkinson's disease vary greatly among individuals, and even the same "gait disturbance" can have different causes and manifestations. While AI-driven personalization is a strength, it also increases the complexity of adjustments.

Second, stability in daily life is crucial. In the laboratory, walking tasks and sensor environments can be controlled to some extent. However, in real life, countless factors such as sudden turns, crowds, steps, fatigue, medication effectiveness, lack of sleep, and stress overlap. The algorithm that reads motor states from brain signals must be able to adjust stimulation safely without malfunctioning in such environments.

Third, the evolution of the device itself is necessary. In this study, adjustments focused on the amplitude of stimulation within the constraints of existing implantable stimulators. In the future, more flexible control may be required, including the frequency of stimulation, the position of electrodes, and approaches to multiple neural circuits. Comprehensive progress is needed, not just in AI, but also in hardware, electrode design, battery, safety management, and programming environments that are easy for doctors to handle.

Nevertheless, this research is significant because it is changing the concept of Parkinson's disease treatment. Traditional treatments have aimed to suppress symptoms as evenly as possible. However, the human body is constantly changing. The necessary brain activity differs when sitting and walking. The condition also changes when medication is effective and when it is wearing off. Therefore, treatment should not be fixed but should change according to the body's changes. This research demonstrated that concept with concrete devices and clinical data.

This concept has the potential to extend beyond Parkinson's disease to other neurological disorders in the future. Technologies that read brain or spinal cord signals and compensate for lost functions are already being researched in fields such as spinal cord injury, epilepsy, depression, and chronic pain. The important point is that AI is not just becoming a "smart diagnostic tool" but is evolving into an entity that adjusts treatment while interacting with the nervous system.

Of course, the more promising the technology, the more caution is required. Since it is a device implanted in the brain, there are many considerations, such as surgical risks, infection, device malfunction, long-term effects, data handling, and patient consent and understanding. Given that brain signals are extremely personal information, discussions on privacy and ethics cannot be avoided. The future of AI treatment depends not only on the performance of the technology but also on whether society can establish a system that patients can use with confidence.

Nevertheless, the reason this research has attracted much attention is clear. For Parkinson's patients, walking is not just a matter of physical ability. Being able to go to the bathroom independently, walk with family, leave the house without fear of falling, and stand up without needing help are deeply connected to independence and dignity.

AI adjusts brain stimulation, and treatment changes with each step the patient takes. What sounded like science fiction a few years ago is now being tested on real patients as clinical research. The results of this study are not yet the goal. However, it marks an important step in Parkinson's disease treatment, moving from merely "suppressing symptoms" to "restoring the movement of life."

If effectiveness and safety are confirmed in large-scale clinical trials in the future, AI-assisted deep brain stimulation may become a new option for patients struggling with gait disturbances. From treatment that suppresses tremors to treatment that supports walking ability, medical care where the brain and AI collaborate is poised to bring hope that was previously difficult to reach in the daily lives of Parkinson's patients.

Source URL

WELT: "“Völlig neue Möglichkeiten” – KI-Therapie lässt Parkinson-Patienten wieder besser laufen." The article that served as the starting point for this article.
https://www.welt.de/gesundheit/plus6a30fbd97e682fc37fbfbdbf/gehirn-voellig-neue-moeglichkeiten-ki-therapie-laesst-parkinson-patienten-wieder-besser-laufen.html

Primary Paper: Nature Medicine "Activity-dependent adaptive deep brain stimulation improves gait in Parkinson’s disease." Used for confirmation of research content, number of subjects, methods, and limitations.
https://www.nature.com/articles/s41591-026-04432-4

EPFL Official Announcement: "When brain stimulation learns to walk with you." Used for confirmation of research overview, real-time stimulation adjustment by AI, and patient comments.
https://actu.epfl.ch/news/when-brain-stimulation-learns-to-walk-with-you/

Neuro X Institute / EPFL Related Announcement: "Adaptive Neuromodulation for Parkinson’s Gait Deficits." Used for confirmation of brain signal decoding of walking states and relationship with past research.
https://neuro-x.epfl.ch/en/news/scientists-decode-the-neural-signals-that-encode-walking-in-the-brain/

SWI swissinfo.ch: "Swiss AI brain ‘pacemaker’ helps Parkinson’s patients walk." Used for confirmation of research announcement, analysis of 35 people, and demonstration test with 4 people, and future challenges.
https://www.swissinfo.ch/eng/swiss-ai/ai-brain-pacemaker-helps-parkinsons-patients-walk/91589796

WHO Fact Sheet "Parkinson disease." Used for confirmation of basic information, symptoms, and global burden of Parkinson's disease.
https://www.who.int/news-room/fact-sheets/detail/parkinson-disease

Parkinson’s Foundation "Deep Brain Stimulation." Used for general explanation of deep brain stimulation therapy and confirmation of challenges in gait and balance.
https://www.parkinson.org/living-with-parkinsons/treatment/surgical-treatment-options/deep-brain-stimulation

LinkedIn: Post introducing the paper by Stefano Scafa. Used for confirmation of SNS dissemination and reactions by the researcher himself.
https://www.linkedin.com/posts/stefano-scafa-aa26a8190_naturemedicine-parkinsonsdisease-neuroscience-activity-7472291364861358080-ST0L

LinkedIn: EPFL School of Engineering official page. Used for confirmation of EPFL's research introduction post and reactions on SNS.
https://www.linkedin.com/company/epfl-school-of-engineering

LinkedIn: CHUV official page. Used for confirmation of CHUV's research introduction post and reactions on SNS.
https://ch.linkedin.com/company/chuv