A cardiac pulse signal affects local field potentials recorded from deep brain stimulation electrodes across clinical targets

This study demonstrates that a cardiac pulse signal, distinct from electrocardiographic artifacts, systematically contaminates local field potential recordings across diverse deep brain stimulation targets and patient groups, with spectral content extending into clinically relevant frequencies that necessitate careful screening for accurate biomarker detection and closed-loop therapy.

The Gist

Imagine your brain is a bustling city, and doctors have installed tiny microphones (called Deep Brain Stimulation electrodes) to listen to the conversations of the neurons. These microphones are crucial for a new type of therapy called "Closed-Loop DBS." Think of this therapy as a smart thermostat for the brain: it listens to the brain's mood, and if it detects a "storm" (like a tremor or a seizure), it automatically sends a calming signal to fix it.

However, there's a problem. Just like a smart thermostat might get confused by a draft from an open window, these brain microphones are picking up a strange, rhythmic noise that isn't part of the brain's conversation at all.

The "Heartbeat Echo"

The paper discovers that these microphones are picking up a pulse signal from the heart.

Usually, we know about heart noise in brain recordings as an electrical "zap" (the ECG QRS complex), which looks like a sharp spike. But this new signal is different. It's not an electrical zap; it's a mechanical thump.

The Analogy: Imagine you are trying to listen to a quiet conversation in a room, but someone is gently tapping a drum on the wall every time their heart beats.

• The tap isn't loud enough to drown out the conversation completely.
• But it creates a rhythmic vibration that shakes the table, making the conversation sound wobbly and distorted.
• Sometimes, the vibration is so subtle you don't even see the table shaking, but the sound is still there.

This "drum tap" is caused by the brain physically pulsing with every heartbeat. As blood rushes into the skull, the brain swells slightly, and the cerebrospinal fluid moves. This physical movement jiggles the tiny wires of the electrode, creating a voltage signal that looks like a heartbeat.

The "Invisible" Noise

The researchers found this "heartbeat echo" in many different parts of the brain (the Parkinson's area, the pain centers, the epilepsy centers).

Here is the tricky part: Commercial medical devices have built-in filters.
Think of these filters like noise-canceling headphones. They are set to block out low, rumbling sounds (like the main "thump" of the heartbeat) so the doctor can hear the higher-pitched brain waves.

• The Problem: The filters block the main thump, but they let the higher harmonics (the subtle vibrations) slip through.
• The Result: The doctor thinks the signal is clean, but it's actually still slightly wobbly. This is dangerous because the "smart thermostat" might mistake this wobble for a brain storm and send an unnecessary shock, or miss a real storm because the noise is hiding it.

The New Detective Tool: "PulsAr"

The authors created a new computer program called PulsAr (Pulsatile Artifact Recognition).

• How it works: Instead of needing a separate heart monitor (ECG) to know when the heart beats, PulsAr listens to the brain signal itself. It looks for a pattern that repeats at the exact speed of a human heartbeat (like a drumbeat).
• The Magic: It can tell the difference between a "clean" brain signal and one that is being shaken by the heart's pulse, even if the shaking is too small for a human eye to see on a screen.

Why This Matters

1. Safety First: If a closed-loop system (the smart thermostat) is running on a signal that is secretly being shaken by the heart, it might make the wrong decisions. This paper warns doctors to check their signals for this "heartbeat echo" before trusting the machine.
2. It's Everywhere: This isn't just a problem in one part of the brain. It happens in the brainstem, the thalamus, and the motor areas. It's a city-wide issue.
3. The "Ghost" in the Machine: In some patients (like those with epilepsy), the brain's own activity is so loud that it hides the heartbeat echo. The computer (PulsAr) can still hear the ghost, even if the human eye can't see it.

The Bottom Line

The brain is a physical object that moves with every heartbeat. This movement creates a "ghost signal" in our brain recordings. While we can't stop the heart from beating, we now have a better way to detect this ghost signal so that our life-saving brain therapies don't get confused by the rhythm of the heart.

In short: The heart is tapping on the brain's microphone. We need to make sure the doctors know it's just a tap, not a scream, so the therapy works correctly.

Technical Summary

1. Problem Statement

Deep Brain Stimulation (DBS) systems increasingly utilize sensing capabilities to record Local Field Potentials (LFPs) for patient monitoring and Closed-Loop DBS (CL-DBS), where stimulation is adjusted based on neural biomarkers. A critical challenge in this domain is artifact contamination, which can lead to incorrect biomarker detection and unsafe therapy adjustments.

While Electrocardiographic (ECG) artifacts (specifically the QRS complex) are well-documented, this paper addresses a less understood phenomenon: a cardiac pulse signal that is distinct from the electrical QRS complex.

• The Issue: This pulsatile signal is temporally aligned with the heartbeat but lacks the characteristic QRS morphology. It is often visually masked by commercial device high-pass filters (typically 1–12 Hz), leading researchers and clinicians to believe the data is clean.
• The Risk: Despite being filtered visually, the signal contains spectral energy extending beyond the fundamental heart rate (up to >10 Hz), potentially overlapping with clinically relevant frequency bands (e.g., delta, theta, beta) and disrupting adaptive control algorithms.
• Knowledge Gap: The prevalence of this signal across different brain targets (beyond the subthalamic nucleus) and patient populations was unknown, and no ECG-independent algorithm existed to detect it automatically.

2. Methodology

The study employed a cross-sectional analysis of prospectively collected LFP data from multiple clinical trials and recording setups.

Data Sources:

• Cranially Mounted System: Picostim™-DyNeuMo system used in three trials:
  • PPN (Pedunculopontine Nucleus): Multiple System Atrophy (MSA) patients ($n=3$).
  • PAG (Periaqueductal Gray) & ST (Sensory Thalamus): Chronic post-stroke pain patients ($n=3$).
  • CMT (Centromedian Thalamic Nucleus): Pediatric Lennox-Gastaut epilepsy patients ($n=3$).
• Externalized System: Medtronic 3389/SenSight leads in the STN (Subthalamic Nucleus) of a Parkinson's Disease patient ($n=1$).
• Total Data: 445 montage sweep LFPs (20-second bipolar recordings) from the Picostim system and 30 STN recordings.

Signal Processing & Algorithm Development (PulsAr):

• Preprocessing: Signals were filtered (0.5 Hz high-pass, 100 Hz low-pass) and downsampled to 1000 Hz.
• PulsAr Algorithm: An ECG-independent detection algorithm was developed using cross-correlation and autocorrelation.
  • It identifies highly repetitive features in the signal.
  • Criteria for Contamination:
    1. Autocorrelation RMS power exceeds a specific threshold.
    2. Inter-peak intervals correspond to physiologic heart rates (50–120 bpm).
    3. Low variability in inter-peak intervals.
  • Template Extraction: Once detected, the algorithm extracts a representative "pulsatile template" by averaging correlated segments, allowing for the creation of a pure artifact string.
• Validation: Two raters visually classified traces as "clean" or "contaminated." The algorithm's performance was validated against this consensus ground truth using ROC analysis.
• Spectral Analysis: Power Spectral Densities (PSD) were computed using Welch's method. Normalized spectral differences were calculated to quantify contamination impact across frequency bands (Delta, Theta, Alpha, Beta, Gamma).

3. Key Contributions

1. Discovery of Prevalence: Demonstrated that cardiac pulse artifacts are not isolated to specific regions but are a brain-wide phenomenon, affecting PPN, PAG, ST, CMT, and STN.
2. PulsAr Algorithm: Introduced a novel, ECG-independent algorithm that can automatically detect and extract this pulsatile artifact without requiring concurrent ECG recordings.
3. Spectral Characterization: Revealed that while the signal is dominated by low frequencies, it possesses significant spectral content extending above 10 Hz, potentially interfering with beta and gamma biomarkers.
4. Morphological Insight: Identified that the artifact morphology resembles Intracranial Pressure (ICP) waveforms (with P1 and P2 peaks) rather than electrical QRS complexes, suggesting a mechanical origin.

4. Key Results

Prevalence:

• The PulsAr algorithm detected contamination in 33.0% of all analyzed LFPs.
• Breakdown by Target:
  • PPN: 53.5% contaminated.
  • PAG: 44.0% contaminated.
  • ST: 22.6% contaminated.
  • CMT: 5.8% contaminated (algorithmically detected, though visually indistinguishable due to high endogenous signal amplitude).
• Contamination was not consistent across all contact pairs or recording sessions, complicating simple mitigation strategies like contact selection.

Spectral Impact:

• PPN & PAG: Significant increases in Delta band power in contaminated vs. clean signals.
• ST: Significant differences in Beta and Gamma bands.
• CMT: Significant decreases in Theta and Alpha power in contaminated traces, suggesting a non-additive interaction where the artifact masks or interacts with endogenous signals.
• STN: No significant differences found in band power, likely due to the small sample size or high background beta activity.

Artifact Characteristics:

• Morphology: The extracted templates showed a primary peak (P1) and often a secondary peak (P2), mirroring ICP waveforms.
• Frequency Content: Power spectra of the pure artifact templates showed dominant power in low frequencies but extended up to 10–15 Hz.
• Stability: The amplitude of the pulsatile signal fluctuated over short timescales (5-second windows), making static subtraction methods difficult.
• Filtering Limitations: Applying high-pass filters (up to 12 Hz) reduced the amplitude of the main pulse but failed to remove higher-frequency components, which remained visible against background neural noise.

5. Significance and Clinical Implications

For Closed-Loop DBS (CL-DBS):

• Safety & Efficacy: The presence of this artifact poses a risk for CL-DBS systems. If the system interprets the pulsatile signal as a biomarker (e.g., a beta burst), it may trigger inappropriate stimulation adjustments.
• Biomarker Reliability: Researchers studying low-frequency oscillations (Delta/Theta) must account for this artifact, as it can artificially inflate power estimates.
• Mitigation Challenges: Simple high-pass filtering is insufficient. The temporal variability of the signal makes linear subtraction difficult without a reference signal (which PulsAr attempts to provide via template extraction).

Physiological Mechanism:

• The authors hypothesize an electromechanical coupling mechanism. Cardiac-driven brain pulsatility (changes in vascular/CSF volume) causes physical deformation of the electrode or relative movement of lead components, generating voltage differences. This is supported by the waveform's similarity to ICP and the lack of QRS morphology.

Future Directions:

• Screening: PulsAr can be used as a screening tool to ensure data quality before biomarker analysis.
• Repurposing: The signal could potentially be used as a proxy for intracranial pressure dynamics or heart rate monitoring (e.g., for seizure detection or sleep staging) if its physiological origin is fully characterized.
• Hardware Design: Future DBS leads may need to be designed to minimize electromechanical coupling to reduce this artifact.

Conclusion:
The paper establishes that cardiac pulse signals are a pervasive, often hidden source of contamination in DBS recordings. It provides a robust tool (PulsAr) to detect this artifact and warns clinicians and researchers that standard filtering may not be enough to ensure the accuracy of neural biomarkers used for adaptive therapy.