Summary: Researchers at Duke University developed a computer algorithm that designs energy-efficient deep brain stimulation patterns. These optimized patterns can maintain symptom relief for Parkinson’s disease while reducing energy use and the frequency of battery replacement surgeries.
Source: Duke University
New method preserves treatment effectiveness while cutting energy use and moving toward personalized therapy for Parkinson’s disease
Duke University biomedical engineers have used an evolutionary computer algorithm to design improved temporal patterns for deep brain stimulation (DBS) used to treat motor symptoms of Parkinson’s disease. By changing the timing of electrical pulses rather than simply using a constant high-frequency stream, the team found patterns that greatly reduce power consumption without sacrificing symptom control.
Deep brain stimulation, first introduced clinically in 1987, delivers electrical pulses to structures within the basal ganglia to improve motor control in people with Parkinson’s disease and other movement disorders. Although DBS is effective for many patients, implanted pulse generators rely on batteries that typically last three to five years and must be surgically replaced, exposing patients to repeated procedures and a cumulative risk of infection.
While investigating how DBS timing affects neural activity, the researchers observed that purely random pulse timing reduced efficacy, suggesting that structured, non-random temporal patterns might outperform continuous high-frequency stimulation. Building on earlier serendipitous discoveries of effective patterns, the team developed a model-based, evolutionary optimization algorithm to search the huge space of possible timing sequences.
The algorithm treats candidate pulse patterns like organisms in a simulated evolutionary process. It begins with a small population of randomly generated patterns and evaluates them in a computational model of Parkinsonian neural activity. Better-performing patterns are more likely to “reproduce,” producing new variants by combining elements of their timing sequences and introducing random mutations. Periodically, new random patterns are injected to maintain diversity. Over thousands of generations, the algorithm converges on temporal patterns that balance therapeutic effectiveness and energy efficiency.
To create a practical search space, the team divided each second of stimulation into five segments and discretized each segment into 200 one-millisecond slices. Each slice either delivered a pulse or remained blank, producing a vast number of potential patterns. The evolutionary method enabled the researchers to explore this space efficiently and identify patterns that dramatically lower pulse counts while preserving clinical benefit.
One evolved pattern delivered an average of only 45 pulses per second—far fewer than the 130 to 185 pulses per second typical of standard high-frequency DBS—resulting in an estimated energy reduction of 60 to 75 percent. Such savings could substantially extend the life of an implanted battery, potentially doubling or tripling the time between replacement surgeries and reducing the cumulative surgical risk for patients implanted at younger ages.
After promising preclinical results in a rodent model of Parkinson’s disease, the researchers collaborated with neurosurgeons at Duke Health and Emory Healthcare to test the optimized pattern in human patients who were undergoing routine battery replacement surgeries. Because many modern DBS devices cannot deliver arbitrary temporal patterns, the team temporarily connected experimental hardware to patients’ implanted leads during the surgical interval when patients were awake under local anesthesia. This allowed clinicians and researchers to directly compare the evolved pattern with each patient’s clinically optimized settings.
The computationally evolved pattern performed comparably to each patient’s long-standing, clinician-tuned stimulation settings while using substantially less energy. Both the optimized pattern and standard high-frequency stimulation suppressed abnormal oscillatory activity in the basal ganglia—an electrophysiological hallmark associated with Parkinsonian motor symptoms—supporting the idea that targeted temporal modulation can achieve the same therapeutic effect with fewer pulses.
Although the precise mechanisms by which abnormal basal ganglia oscillations generate Parkinsonian symptoms—and by which DBS interrupts those pathological rhythms—remain areas of active research, this study demonstrates that model-based computational evolution can identify clinically viable stimulation strategies that increase energy efficiency and open the door to personalized temporal patterns tailored to a patient’s symptom profile.
Funding: This research was supported by the National Institutes of Health, including a Javits Neuroscience Investigator Award to Warren M. Grill (R01-NS040894, R37-NS040894, R01-NS079312).
Source: Ken Kingery – Duke University
Image credit: Duke University
Abstract
Optimized temporal pattern of brain stimulation designed by computational evolution
Brain stimulation is a promising therapy for several neurological disorders, including Parkinson’s disease. Traditional stimulation parameters focus on frequency and amplitude. In this work, researchers varied the temporal pattern of deep brain stimulation and used model-based computational evolution to optimize that pattern for symptom relief. The optimized pattern provided symptom reduction comparable to standard high-frequency stimulation (130–185 Hz) and performed better than frequency-matched standard stimulation in a rodent model and in human patients. Both optimized and standard stimulation suppressed abnormal basal ganglia oscillations. These results demonstrate that computational evolution of temporal patterns can increase the efficiency of brain stimulation, reducing the energy required for effective treatment compared to current paradigms.
Reference: “Optimized temporal pattern of brain stimulation designed by computational evolution” by David T. Brocker, Brandon D. Swan, Rosa Q. So, Dennis A. Turner, Robert E. Gross, and Warren M. Grill. Science Translational Medicine. Published online January 4, 2017. doi:10.1126/scitranslmed.aah3532