Precision phase targeting of event-related oscillations using real-time closed-loop TMS-EEG

This paper introduces and validates a real-time closed-loop TMS-EEG system that directly detects oscillatory phase without prediction, demonstrating superior triggering probability and phase precision compared to traditional prediction-based methods for targeting both spontaneous and event-related brain oscillations during active cognition.

The Gist

Imagine your brain is like a bustling orchestra. The musicians (neurons) are constantly playing rhythms and melodies (brain waves) that change depending on what you're doing—whether you're solving a puzzle, remembering a route, or just sitting with your eyes closed.

For years, scientists have wanted to use a special tool called TMS (Transcranial Magnetic Stimulation) to "conduct" this orchestra. TMS is like a gentle, non-invasive magnetic tap on the head that can nudge the brain's activity. But there's a catch: to get the best effect, you have to tap the musicians at the exact right moment in their rhythm. If you tap when they are resting, nothing happens. If you tap when they are in the middle of a note, you might ruin the song.

The Old Way: Guessing the Future

The traditional way to do this is called Phase Prediction. Imagine you are trying to hit a moving target, like a ball thrown in the air. Since there is a tiny delay between when you see the ball and when your hand moves, you have to guess where the ball will be a split second from now.

Current brain-computer systems work the same way. They look at the brain waves from the last few seconds, try to predict what the rhythm will be a millisecond later, and then send the TMS tap.

• The Problem: This works great if the rhythm is steady and predictable, like a metronome ticking away while you sleep. But when you are actively thinking, solving a maze, or remembering something, the brain's rhythm is chaotic. It starts and stops suddenly. Predicting the future of a chaotic rhythm is like trying to guess the next move of a jazz drummer who is improvising. The prediction often misses the mark.

The New Way: The "Instant Reflex"

The authors of this paper built a brand-new system called RT-CL (Real-Time Closed-Loop). Instead of guessing the future, this system acts like a lightning-fast reflex.

Think of it this way:

• The Old System (Prediction): A human conductor looking at a score, calculating the next beat, and then waving the baton. By the time they wave, the beat has already passed.
• The New System (RT-CL): A robot conductor with super-speed reflexes. It doesn't calculate; it sees the beat happen and taps the drum instantly.

They achieved this by using a special computer chip called an FPGA. While normal computers take a few milliseconds to process data (which is an eternity in brain time), this chip processes data in microseconds. It's so fast that it can detect the exact moment a brain wave hits a specific point (like the peak of a wave) and fire the TMS pulse almost immediately, without needing to guess.

What They Tested

To prove their new system worked, they ran three tests:

1. The Chirp Test: They fed a computer a sound that slowly changed pitch (like a bird singing from a high note to a low note). The new system hit the target almost 100% of the time, while the old guessing system missed often, especially when the pitch changed quickly.
2. The "Eyes Closed" Test: They asked people to sit with their eyes open and closed. This creates a steady, predictable brain rhythm (Alpha waves). Both systems worked okay here, but the new system was still more accurate and hit the rhythm more often.
3. The "Maze" Test (The Big Challenge): This was the real test. They asked people to play a virtual maze game and a memory game. These tasks create Event-Related Oscillations (EROs). These are brain bursts that happen only when you get a reward or make a decision. They are short, messy, and unpredictable.
   • The Result: The old "guessing" system struggled badly here, missing many opportunities and hitting the wrong times. The new RT-CL system was a superstar. It caught about 18% more of these fleeting brain bursts and hit the exact timing 19 degrees more accurately than the old system.

Why This Matters

Why should you care?

1. Better Science: Scientists can finally study how the brain works while we are doing complex things like thinking or feeling, not just while we are resting. They can now ask, "What happens if we tap the brain exactly when you remember a face?"
2. Better Medicine: Currently, treatments for depression or OCD using magnetic stimulation are like taking a pill at the same time every day, regardless of how you feel. This new system allows doctors to deliver treatment only when the patient's brain is in a specific "state" (like when they are feeling anxious or remembering a trauma). It's the difference between giving a patient a pill randomly versus giving it the exact second their brain needs it most.

The Bottom Line

The authors have built a "smart conductor" for the brain. Instead of guessing where the music is going, it listens to the music as it happens and conducts in perfect sync. This opens the door to a new era of precision brain science and treatments that are tailored to the exact moment your brain needs help.

Technical Summary

1. The Problem

Current closed-loop Transcranial Magnetic Stimulation (TMS) systems combined with EEG (TMS-EEG) rely on phase prediction algorithms (PPA). These algorithms estimate the future phase of neural oscillations based on hundreds of milliseconds of preceding data.

• Limitations: Prediction methods require highly periodic, stable oscillations (e.g., spontaneous resting-state rhythms). They fail when targeting transient, event-related oscillations (EROs) that emerge briefly (few cycles) following sensory or cognitive events.
• Latency Issues: Standard software-based processing (data streaming, algorithm execution on PCs) introduces variable delays (tens of milliseconds) and phase distortions (due to causal filtering). To compensate, systems must "predict" the phase forward, which introduces errors, especially when the signal undergoes abrupt phase resets or amplitude fluctuations common in active cognition.
• Consequence: This limits the ability to probe causal relationships between neural phase and cognitive processes in real-time and hinders the development of precision interventions for pathological brain states during active tasks.

2. Methodology

The authors developed and validated a Real-Time Closed-Loop (RT-CL) TMS-EEG system that bypasses prediction in favor of direct, microsecond-scale phase detection.

• Hardware Architecture:

  • FPGA Processing: The system utilizes Field Programmable Gate Arrays (FPGAs) (National Instruments USB-7845R with Xilinx Kintex-7) to process analog EEG signals directly. This eliminates serial processing bottlenecks and reduces latency to the microsecond scale.
  • Signal Chain: Analog EEG signals are captured via an EEG8 amplifier, filtered (Butterworth filters), and processed on the FPGA. The system streams data to a PC for monitoring but triggers TMS pulses based on FPGA logic.
  • Algorithm: The FPGA algorithm consists of four modules:
    1. Filter Bank: Parallel 2nd/4th order Butterworth bandpass filters (e.g., 4–10 Hz for theta, 9–12 Hz for alpha).
    2. Phase Detection: Identifies peaks, troughs, and zero-crossings using the signal and its derivative.
    3. Amplitude Thresholding: Selects the frequency band with the highest power to ensure robust detection.
    4. Triggering Logic: Sends a trigger only when the target phase is detected, amplitude exceeds a threshold, and an event gate (task feedback) is active.

• Experimental Validation:

  • Experiment 1 (Simulated Signal): A frequency-modulated "chirp" signal (50 Hz to 1 Hz) was fed into both the RT-CL system and a standard Phase Prediction Algorithm (PPA-CL) running offline on recorded data.
  • Experiment 2 (Human EEG): 18 participants underwent three conditions (no TMS delivery was used to isolate system performance):
    1. Spontaneous Alpha: Eyes-open vs. eyes-closed resting states (targeting occipital alpha, 9–12 Hz).
    2. T-Maze Task: A spatial navigation task eliciting Right Posterior Theta (RPT, 4–10 Hz) upon feedback.
    3. Linear Track Memory (LTM) Task: A continuous movement task eliciting RPT, characterized by higher trial-to-trial variability.
  • Comparison: The RT-CL system was compared against the PPA-CL system (implemented in EventIDE software) using identical parameters and calibration periods.

3. Key Contributions

• Novel Hardware/Software Integration: Demonstrated the first successful adaptation of an FPGA-based, microsecond-latency closed-loop system for non-invasive human TMS-EEG targeting transient oscillations.
• Direct Detection vs. Prediction: Proved that direct phase detection is superior to phase prediction for non-stationary, event-related signals.
• Event Gating: Implemented a logic gate that restricts triggering to specific task-relevant time windows (e.g., 50ms before to 450ms after feedback), ensuring stimulation occurs only during relevant cognitive processing.

4. Results

The RT-CL system outperformed the PPA-CL system across all metrics (triggering probability/sensitivity and phase error precision/SDPE).

| Metric | Condition | RT-CL Performance | PPA-CL Performance | Improvement |
| :--- | :--- | :--- | :--- :--- |
| Triggering Probability | Simulated Chirp | 90–93% (consistent across 3.5–10 Hz) | ~73–77% (dropped <10% at 10 Hz) | +16–24% |
| Phase Error (SDPE) | Simulated Chirp | 7.0° – 9.7° | 9.4° – 11.6° | **19% higher precision** |
| Triggering Probability | Spontaneous Alpha | 70% | ~59% | +11% |
| Phase Error (SDPE) | Spontaneous Alpha | 29.7° | 40.1° | **10° lower error** |
| Triggering Probability | T-Maze (RPT) | 83% | 72% | +11% |
| Phase Error (SDPE) | T-Maze (RPT) | 42.6° | 55.6° | ~13° lower error |
| Triggering Probability | LTM (RPT) | 85.1% | 67% | +18% |
| Phase Error (SDPE) | LTM (RPT) | 41.6° | 60.3° | ~19° lower error |

• Timing Behavior: The PPA-CL system consistently triggered before the target phase (negative phase error) due to prediction extrapolation, whereas the RT-CL system triggered slightly after the target phase (positive error) as designed by direct detection.
• Robustness: The performance gap widened significantly in the LTM task, where the RPT signal was less stereotyped and more variable, highlighting PPA-CL's failure to handle non-stationary signals.

5. Significance

• Mechanistic Neuroscience: This technology enables the causal investigation of phase-dependent neural mechanisms during active cognition (e.g., memory encoding, decision-making), which was previously impossible due to the transient nature of the signals.
• Precision Neuromodulation: It paves the way for "state-and-context-dependent" clinical interventions. Unlike current rTMS protocols that use fixed frequencies, RT-CL can target pathological brain states as they emerge during symptom provocation or specific cognitive tasks.
• Overcoming Prediction Limits: The study establishes that for signals with phase resets (common in EROs), prediction is fundamentally flawed. Direct detection via FPGA is the necessary approach for high-precision targeting of gamma bursts and event-related theta.
• Future Applications: The system is a platform for future studies on phase-amplitude coupling (gamma), complex therapeutic protocols, and multi-modal integration (MRI/physiology) for next-generation psychiatry and neurology treatments.