Multimodal Biomarker-Guided Deep Brain Stimulation Programming in Parkinson's Disease: The DBSgram Framework

This paper introduces DBSgram, a multimodal framework that integrates subthalamic nucleus local field potentials and wearable kinematic data to enable objective, data-driven Deep Brain Stimulation programming for Parkinson's disease by visualizing the relationship between stimulation parameters, neural activity, and motor symptom improvement.

The Gist

Imagine you are trying to tune a very old, complex radio to find a clear station. In the past, doctors treating Parkinson's disease with Deep Brain Stimulation (DBS) had to do something similar, but with a major catch: they were tuning the radio while blindfolded.

They would adjust the electrical current, ask the patient, "Does your hand feel less stiff?" or "Is your tremor gone?" and guess the best setting based on how the patient felt at that exact moment. It was a slow, subjective process, like trying to find the perfect volume on a radio by guessing rather than looking at a signal meter.

This paper introduces a new tool called DBSgram that takes off the blindfold. It turns the "guessing game" into a precise, data-driven science by combining two different types of information into one easy-to-read dashboard.

Here is how it works, broken down into simple concepts:

1. The Two "Ears" of the System

To tune the brain correctly, the system listens to two different things at the same time:

• The Brain's "Radio Static" (LFP): The implanted device in the patient's brain can "listen" to the brain cells in the area causing the Parkinson's symptoms. When the disease is active, these cells chatter in a specific, annoying rhythm (called "beta waves"). Think of this as a loud, staticky hiss on the radio that drowns out the music.
• The Body's "Motion Sensors" (Wearables): The patient wears small, smart sensors on their hands (like high-tech fitness trackers). These sensors measure exactly how much the hand is shaking, how stiff the wrist is, or how slow the movements are. This is like a camera recording the radio's volume knob turning.

2. The "Time-Travel" Problem

The biggest challenge was that the brain implant and the wrist sensors are two different devices with their own internal clocks. It's like trying to match a video recording from a phone with a sound recording from a tape recorder; if they aren't perfectly synced, the audio and video are out of step.

The researchers solved this with a clever trick: The "Clap" Sync.
At the start of the session, the patient taps the device under their skin. This creates a sharp "clap" that is instantly recorded by both the brain implant and the wrist sensors. The computer uses this "clap" as a starting line to align the two data streams perfectly, so they know exactly what the brain was doing at the exact moment the hand moved.

3. The "DBSgram" Dashboard

Once the data is synced, the system creates a visual report called the DBSgram.

Think of this like a GPS for the brain.

• The Map: The horizontal axis shows how much electricity (amplitude) the doctor is sending.
• The Traffic: The vertical lines show two things happening at once:
  1. The Static: How much the "beta noise" in the brain is going down.
  2. The Movement: How much the hand tremor or stiffness is going down.

4. Finding the "Sweet Spot"

In the past, a doctor might turn up the volume until the tremor stopped, but they wouldn't know if they were too close to the edge where side effects (like muscle twitching or speech problems) would start.

With the DBSgram, the doctor can see the "Therapeutic Window" clearly:

• The Ideal Responder: Imagine a patient where the "static" drops and the "movement" gets smooth at the exact same time. The dashboard shows a wide, green zone. The doctor knows, "Perfect! We can set the volume anywhere in this green zone."
• The Complex Case: Imagine a patient where the brain stops "chattering" (static drops), but the hand doesn't get better until the volume is turned way up. However, turning it up that high causes side effects. The dashboard shows this conflict clearly. It tells the doctor: "Hey, the brain is ready, but the body isn't responding until we push harder. We need to be very precise here, maybe use a special 'directional' antenna to aim the electricity more narrowly to avoid the side effects."

Why This Matters

This study tested the system on 18 patients. While only a few had "perfect" data (because real life is messy and patients sometimes move too much), the results were promising.

• It's Objective: It removes the guesswork and the "I think it feels better" subjectivity.
• It's Fast: It helps doctors find the right settings in a standard 40-minute appointment rather than hours of trial and error.
• It's Personal: Every patient's brain is different. This tool creates a custom map for each person, ensuring they get the exact right dose of electricity.

In short: The DBSgram is like giving the doctor a pair of glasses that lets them see the invisible electrical signals of the brain and the physical movements of the body on the same screen. Instead of guessing the right setting, they can simply look at the dashboard and drive straight to the "Sweet Spot."

Technical Summary

1. Problem Statement

Deep Brain Stimulation (DBS) is a standard therapy for Parkinson's Disease (PD), but optimizing stimulation parameters (amplitude, frequency, pulse width) remains a significant clinical bottleneck. Current programming relies on:

• Subjective Assessment: Clinicians depend on the Unified Parkinson's Disease Rating Scale (UPDRS), which is prone to inter-rater variability and patient fatigue.
• Time-Consuming Process: Iterative "trial-and-error" adjustments often take hours, leading to suboptimal settings or delayed identification of side effects.
• Data Silos: While sensing-enabled neurostimulators can record Local Field Potentials (LFPs) and wearable sensors can track kinematics, these data streams are typically acquired on independent hardware with unsynchronized clocks, making it difficult to correlate neural activity with motor performance in real-time.

2. Methodology

The authors propose DBSgram, a multimodal framework that integrates neurophysiological and kinematic data into a unified visualization tool. The methodology consists of five primary stages:

A. Hardware and Participants

• Cohort: 18 patients with idiopathic PD implanted with Medtronic Percept™ PC/RC neurostimulators (sensing-enabled, directional leads).
• Neurophysiology: STN-LFP recordings acquired at 250 Hz via the implant's Brainsense™ Streaming mode.
• Kinematics: Bilateral hand movements recorded using the iHandU wearable system, consisting of four 9-axis IMUs (accelerometer, gyroscope, magnetometer) per session (two per hand).
• Protocol: A standardized 40-minute clinical titration session involving:
  • Mechanical Artifact: Patients tapped the IPG to create a synchronized marker across all data streams.
  • Baseline: Recording with stimulation OFF.
  • Titration: Unilateral amplitude increments (0 mA to max) in 1-minute steps for specific tasks: resting tremor, wrist rigidity, and hand open-close (bradykinesia).
  • Final Baseline: Recording with stimulation ON at clinical settings.

B. Multimodal Synchronization Pipeline

To align independent hardware clocks, a two-stage synchronization strategy was implemented:

1. Coarse Alignment: Network Time Protocol (NTP) server synchronization reduced initial error to < 2.1 seconds.
2. Fine Alignment: A mechanical artifact (tapping the device) generated simultaneous markers in the IMU accelerometer, LFP, and video feed. Offline alignment of these markers reduced the final synchronization error to ≤ 560 ms.

C. Automated Signal Processing

• LFP Processing: Beta-band power (13–35 Hz) was extracted using FFT on 3-second epochs. Power was normalized to the median beta power at 0 mA stimulation to quantify suppression.
• IMU Processing:
  • Tremor: Band-pass filtered (2–8 Hz) accelerometry analyzed via a decision tree based on spectral power ratios to distinguish tremor from rest/voluntary movement.
  • Rigidity: Angular velocity from gyroscopes fed into a pre-trained computational model (validated on 45 patients) to output a rigidity improvement percentage.
  • Bradykinesia: Analysis of repetitive hand open-close cycles to extract total signal power and the slope of angular velocity (indicating motor fatigue).

D. Visualization (DBSgram)

The processed features were integrated into a visual dashboard mapping Stimulation Amplitude (X-axis) against Beta Power and Motor Metrics (Y-axes). This allows clinicians to visually identify the "therapeutic window" where beta suppression correlates with motor improvement without side effects.

3. Key Contributions

1. Synchronized Multimodal Acquisition: Development of a robust pipeline to align implantable LFP data with wearable IMU data with sub-second accuracy during routine clinical workflows.
2. Automated Biomarker Extraction: Creation of symptom-specific algorithms to objectively quantify tremor, rigidity, and bradykinesia from raw sensor data, removing reliance on subjective scoring.
3. Unified Visualization Framework: The DBSgram tool provides a single interface to visualize the dose-dependent relationship between stimulation amplitude, neural biomarkers (beta suppression), and motor outcomes, facilitating the identification of patient-specific therapeutic windows.

4. Results

The framework was validated on a subset of 4 patients (8 hemispheres) after rigorous data quality filtering (excluding data with severe artifacts or incomplete titration).

• Technical Feasibility: The system successfully acquired and synchronized data without disrupting the 40-minute clinical workflow.
• "Ideal Responder" Cases: In patients like Patient 007 (Left Hemisphere) and Patient 014 (Right Hemisphere), the DBSgram showed a strong, dose-dependent correlation: as stimulation amplitude increased, beta power decreased, and motor metrics (UPDRS scores and kinematic measures) improved. This confirmed a clear, stable therapeutic window.
• Complex/Asymmetric Cases:
  • Patient 007 (Right Hemisphere): The highest amplitude (2.0 mA) provided maximum motor improvement but caused capsular side effects. The DBSgram confirmed the physiological efficacy of this level, guiding the clinician to use directional steering to redistribute the field, preserving motor benefits while mitigating side effects.
  • Patient 014 (Left Hemisphere): Showed a "delayed response." Beta power decreased at lower amplitudes, but motor improvement only occurred at the highest level (4.4 mA), which also triggered side effects. The visualization highlighted a "missing" therapeutic window, alerting the clinician that high-resolution titration (fractional mA steps) between 3.3 mA and 4.4 mA was necessary to find a safe setting.

5. Significance and Conclusion

• Objective Programming: DBSgram shifts DBS programming from subjective, time-consuming trial-and-error to a data-driven, objective process.
• Clinical Utility: It acts as a "therapeutic window detector," helping clinicians quickly identify optimal settings, justify advanced interventions (like directional steering), and detect complex asymmetries between hemispheres.
• Future Impact: While currently limited by strict data quality filters and a small proof-of-concept cohort, the framework lays the groundwork for adaptive DBS (aDBS) and closed-loop neuromodulation. By establishing the correlation between neural biomarkers and motor outcomes, it paves the way for automated, personalized therapy adjustments in future clinical settings.

Limitations Noted: The study required strict data curation, excluding many datasets due to artifacts. Future work must improve artifact rejection algorithms and validate the system on a larger, more diverse population to ensure robustness in real-world clinical noise.