Optimal closed-loop deep brain stimulation using multiple independently controlled contacts

Gihan Weerasinghe, Benoit Duchet, Christian Bick, Rafal BogaczPLOS Computational Biology 2021 · DOI

TL;DR

This paper introduces "Adaptive Coordinated Desynchronization" (ACD), a novel closed-loop deep brain stimulation (DBS) strategy that leverages multiple independently controlled electrode contacts to optimally suppress pathological brain oscillations. ACD demonstrates superior efficacy compared to existing methods like coordinated reset (CR) and phase-locked (PL) stimulation, particularly for systems with specific neural response characteristics.

What Problem Does This Paper Address?

Deep brain stimulation (DBS) is a crucial treatment for neurological disorders like Parkinson's disease and essential tremor, which are characterized by pathological synchronous neural activity. While traditional DBS uses open-loop stimulation with fixed parameters, newer multi-contact electrodes offer the potential for more precise and adaptive control. However, the increased complexity of these systems necessitates a better understanding of how to optimally apply stimulation across multiple contacts to achieve therapeutic goals, primarily symptom reduction through desynchronization. This paper tackles the challenge of designing an optimal closed-loop DBS strategy that can effectively utilize these multi-contact electrodes to maximize symptom suppression by maximally desynchronizing neural populations.

Methods

The authors propose Adaptive Coordinated Desynchronization (ACD), a closed-loop strategy that builds upon existing models of coupled oscillators (specifically, the Kuramoto model). ACD aims to determine the optimal stimulation parameters (intensity and timing) for each individual contact to maximally reduce the global amplitude of neural oscillations, which is correlated with symptom severity.

Key methodological aspects include:

  • Multi-population Kuramoto Model: The brain is modeled as a system of coupled neural units organized into multiple populations, each exhibiting oscillatory behavior.
  • Analytical Derivations: The paper derives analytical expressions for the amplitude and phase response of these neural populations to stimulation, considering both global and local factors.
  • Closed-Loop Strategy: ACD stimulates only when the predicted change in global amplitude is negative (i.e., when stimulation is expected to reduce symptoms). The strategy leverages local population amplitudes and phases, along with the global phase, to guide stimulation.
  • Numerical Simulations: The proposed ACD strategy is compared against established methods like Coordinated Reset (CR) and Phase-Locked (PL) stimulation using numerical simulations of the multi-population Kuramoto model.
  • Parameter Space Exploration: Simulations explore the impact of various parameters, including noise levels, coupling constants, electrode-population geometry, and the characteristics of the neural units' response curves (uPRC), on the efficacy of different stimulation strategies.

Datasets

The paper utilizes tremor data from a study by Cagnan et al. [17] to inform the parameters of the multi-population Kuramoto model, specifically for fitting the natural frequency distribution of the neural units to reproduce tremor-like oscillations. The data was filtered and processed as described in the original study.

Key Insights and Ideas

  • Local vs. Global Control: The efficacy of DBS is not solely dependent on global brain activity but is significantly influenced by local population dynamics (amplitude and phase). ACD leverages this by considering both.
  • The Role of the uPRC: The "zeroth harmonic" ($a_0$) of the unit phase response curve (uPRC) is a critical determinant of ACD's utility. Type I uPRCs (large $a_0$) are predicted to benefit more from ACD than Type II uPRCs (small $a_0$).
  • Geometric Dependence: The effectiveness of multi-contact DBS is influenced by the spatial arrangement of electrodes and neural populations. Delocalized stimulation (larger $\eta$) can reduce the advantage of ACD over simpler methods.
  • ACD's Adaptive Nature: Unlike CR, which uses a fixed stimulation pattern, ACD dynamically adjusts stimulation based on the real-time state of the neural system, leading to potentially more efficient and effective desynchronization.
  • Potential for Reduced Stimulation Intensity: ACD shows promise in achieving similar or better desynchronization with lower stimulation intensities compared to CR, which could translate to reduced energy consumption and fewer side effects.

Key Results

  • Superior Efficacy: Numerical simulations demonstrate that ACD achieves greater desynchronization (lower average amplitude $\bar{p}$) compared to PL and CR stimulation across various parameter settings, particularly for systems with Type I uPRCs.
  • Utility of ACD: The utility of ACD over PL stimulation is predicted to be higher for Type I uPRCs and when neural populations are more clustered (larger $k_{diag}$).
  • Energy Efficiency: For comparable efficacy, ACD can achieve its goals with lower stimulation intensities and thus lower total energy consumption than CR, especially at lower stimulation levels.
  • Parameter Sensitivity: The performance of ACD is sensitive to the uPRC characteristics, electrode-population geometry, and the degree of neural clustering.

Limitations and Future Work

The authors acknowledge several limitations:

  • Assumptions of the Model: The model relies on the Ott-Antonsen ansatz for infinite Kuramoto systems, which may not perfectly capture the dynamics of finite, noisy, or more complexly coupled systems.
  • Simplifications in Stimulation Modeling: The model assumes instantaneous effects of stimulation and uses simplified electrode models. Real stimulation pulses have finite durations and shapes.
  • "Small Populations" Assumption: The assumption of small, localized neural populations simplifies electrostatic modeling but might not hold for all systems.
  • Stimulation Artefacts: The current model does not explicitly account for stimulation artefacts in LFP recordings, which could pose a challenge for real-time closed-loop control.
  • Parameter Estimation: Estimating the model's parameters from real patient data and handling uncertainty in these parameters are crucial for real-world application.

Future work suggested includes:

  • Testing on More Realistic Systems: Investigating ACD's robustness with more complex coupling functions, different natural frequency distributions, and higher harmonics in the uPRC.
  • Addressing Stimulation Artefacts: Developing methods to suppress or account for stimulation artefacts in recordings.
  • Real-World Implementation: Bridging the gap between simulations and experimental paradigms by treating the system as a "virtual patient" and estimating parameters from simulated electrode measurements.
  • Exploring Other Perturbation Methods: Investigating the applicability of ACD principles to non-electrode-based stimulation techniques like optogenetics.

Summary updated

BibTeX
@article{weerasinghe2021optimal,
  author = {Gihan Weerasinghe and Benoit Duchet and Christian Bick and Rafal Bogacz},
  title = {{Optimal closed-loop deep brain stimulation using multiple independently controlled contacts}},
  journal = {PLOS Computational Biology},
  year = {2021},
  doi = {10.1371/journal.pcbi.1009281},
}
Figure