A high performing brain–machine interface driven by low-frequency local field potentials alone and together with spikes

Sergey D. Stavisky, Jonathan C. Kao, Paul Nuyujukian, Stephen I. Ryu, Krishna V. ShenoyJournal of Neural Engineering 2015 · DOI

Abstract

Objective. Brain-machine interfaces (BMIs) seek to enable people with movement disabilities to directly control prosthetic systems with their neural activity. Current high performance BMIs are driven by action potentials (spikes), but access to this signal often diminishes as sensors degrade over time. Decoding local field potentials (LFPs) as an alternative or complementary BMI control signal may improve performance when there is a paucity of spike signals. To date only a small handful of LFP decoding methods have been tested online; there remains a need to test different LFP decoding approaches and improve LFP-driven performance. There has also not been a reported demonstration of a hybrid BMI that decodes kinematics from both LFP and spikes. Here we first evaluate a BMI driven by the local motor potential (LMP), a low-pass filtered time-domain LFP amplitude feature. We then combine decoding of both LMP and spikes to implement a hybrid BMI. Approach. Spikes and LFP were recorded from two macaques implanted with multielectrode arrays in primary and premotor cortex while they performed a reaching task. We then evaluated closed-loop BMI control using biomimetic decoders driven by LMP, spikes, or both signals together. Main Results. LMP decoding enabled quick and accurate cursor control which surpassed previously reported LFP BMI performance. Hybrid decoding of both spikes and LMP improved performance when spikes signal quality was mediocre to poor. Significance. These findings show that LMP is an effective BMI control signal which requires minimal power to extract and can substitute for or augment impoverished spikes signals. Use of this signal may lengthen the useful lifespan of BMIs and is therefore an important step towards clinically viable BMIs.

Introduction

The "gold standard" for high-performance brain-machine interfaces (BMIs) has long been the detection of spikes—the individual action potentials of single neurons. However, spikes are notoriously fickle. Because electrodes are foreign objects, the brain’s immune response often creates scar tissue (gliosis) that pushes neurons away from the sensors. Over months or years, the signal-to-noise ratio of these individual spikes often degrades until they disappear entirely.

To make BMIs clinically viable for a lifetime, researchers need signals that are more robust. Local Field Potentials (LFPs)—which represent the aggregate electrical activity of thousands of neurons near the electrode—persist much longer than spikes. Historically, however, LFP-driven BMIs have been slower and less accurate than their spike-based counterparts. This study investigates whether a specific LFP feature, the Local Motor Potential (LMP), can close that performance gap and provide a vital "safety net" for neural control.

Methods

The researchers worked with two rhesus macaques (Monkey R and Monkey J) implanted with 96-channel electrode arrays in the primary motor (M1) and dorsal premotor (PMd) cortices. The technical core of the system was the use of the LMP, defined as the low-pass filtered time-domain amplitude of the LFP (below 500 Hz).

Key methodological innovations included:

  • Half-wave Rectification: The team discovered that "rectifying" the signal (zeroing out positive voltages) prevented a "spring-back" effect where the cursor would unintentionally bounce after a movement.
  • Velocity Kalman Filters: They used a standard probabilistic decoder to map neural activity to cursor velocity. The state-space model for this decoder is defined by the kinematics state update and the neural observation equations.
The Kalman Filter framework used to map neural observations (y) to kinematic states (x).
  • Hybrid Decoding: A "stacked" decoder was implemented to combine both spikes and LMP features into a single control vector.
  • Channel Dropping: To simulate the natural degradation of an implant, the researchers systematically "turned off" the best-performing spike channels to evaluate if the LMP could rescue system performance.

Datasets

The study utilized several datasets recorded from the two macaques:

  • Radial 8 Task: A center-out reaching task used primarily for training the decoders.
  • Continuous Random Target Task: A more challenging evaluation task where the cursor must be moved to targets appearing at random locations.
  • Sedated Control: A crucial control dataset where the monkey was sedated while an experimenter moved its arm. This confirmed that the LMP signal was truly neural and not an electrical artifact caused by physical movement.

Key Results and Insights

Performance and Accuracy

Using LMP alone, Monkey R achieved a 99% success rate on the random target task, significantly outperforming previously reported LFP-based BMIs. Interestingly, the researchers found that the time-domain amplitude (LMP) was a much better predictor of movement than the power in specific frequency bands, suggesting that phase information in low-frequency brain waves is critical for fine motor control.

Hybrid Gains and Robustness

For Monkey R, who had "mediocre" spike quality, adding LMP to the spike decoder improved performance by 12%. In the "channel-dropping" simulation, the hybrid decoder remained functional even after 80% of the spike-carrying channels were removed. While the spikes-only decoder eventually failed completely, the hybrid system "failed gracefully," reverting to the performance level of the LMP signal.

Stability and Control Lag

The study identified that LMP has a longer "control lag" (~100ms) compared to spikes (~50ms). This makes LMP-only control feel slightly more ballistic—effective for reaching targets quickly, but more difficult for fine corrections. However, a major advantage was stability: an LMP decoder trained on Day 1 could still be used effectively three weeks later without any recalibration, addressing a major hurdle for clinical BMIs that typically require daily "tune-ups."

Discussion

The authors acknowledge that if a patient has access to high-quality, "pristine" spikes, adding LFP doesn't offer a significant boost and can even slightly degrade performance due to added signal noise and longer lag.

Future work will likely focus on more sophisticated ways to weight these signals. For instance, a "smart" decoder could automatically rely more on spikes for fine-grained movements near a target and shift toward LMP for high-velocity movements across the screen. Testing this in human clinical trials with participants who have tetraplegia is the next necessary step toward a commercial medical device that remains functional over the long term.

Summary updated

Abstract

These findings show that LMP is an effective BMI control signal which requires minimal power to extract and can substitute for or augment impoverished spikes signals. Use of this signal may lengthen the useful lifespan of BMIs and is therefore an important step towards clinically viable BMIs.
BibTeX
@article{stavisky2015high,
  author = {Sergey D. Stavisky and Jonathan C. Kao and Paul Nuyujukian and Stephen I. Ryu and Krishna V. Shenoy},
  title = {{A high performing brain–machine interface driven by low-frequency local field potentials alone and together with spikes}},
  journal = {Journal of Neural Engineering},
  year = {2015},
  doi = {10.1088/1741-2560/12/3/036009},
}

Loading related content...

Figure