Personalized, EEG-controlled intermittent theta burst stimulation

Dear Editor,

Brain state-controlled and closed-loop approaches to transcranial magnetic stimulation (TMS) have emerged as a promising frontier in precision psychiatry [1]. Multiple groups have demonstrated that the phase of ongoing electroencephalographic (EEG) oscillations modulates cortical susceptibility to a standard TMS pulse, with most evidence to date coming from recordings of evoked potentials in the motor cortex [2]. These findings raise the prospect that EEG-synchronized stimulation could reduce inter-individual variability in therapeutic neuromodulation. However, most EEG phase-informed protocols to date focused on immediate effects using single-pulse or single-burst TMS [2], while few aimed to induce the long-lasting plasticity needed clinically. A prime clinical protocol is intermittent theta burst stimulation (iTBS), FDA-cleared for depression treatment, which delivers 10 bursts (triplets) at 5 Hz over a 2-second train every 10 seconds [3]. Although the protocol's burst rate was designed to mimic endogenous theta rhythms, synchronization of uninterrupted iTBS to ongoing EEG theta remains an open challenge. Two obstacles stand in the way. First, the rapid succession of stimulation pulses within each train overwhelms EEG recordings with sequential artifacts, degrading real-time phase tracking after the first pulse. Second, prefrontal theta oscillations are less stationary and of lower amplitude than the motor alpha rhythms typically targeted in prior closed-loop work, demanding more robust [4] and computationally expensive methods. Overcoming these constraints would enable investigation of the clinical potential of state-controlled iTBS.

Here we present our method for EEG state-controlled iTBS (iTBS-EEG). We demonstrate its technical validity in the prefrontal cortex, enabled by a fast, parallelized system architecture and hierarchical optimization of individual state-control parameters. We also discuss the mechanistic rationale that underlies this method, which we term “seed-and-sustain.”

The core concept is to synchronize iTBS delivery to the individual theta oscillation, which originally inspired the protocol [3]. This achieves two goals: i. matching the inter-burst rate to the patient's individual theta rhythm, and ii. delivering all pulses at the oscillatory phase most susceptible to perturbation. The former has shown promise in open-loop delivery [5]. The latter has preliminary support from a single-burst EEG-controlled stimulation study [6].

The method is described in Figure 1A–B and in supplementary materials. The main challenge is forecasting the biomarker (theta phase), which is needed because of technical delays, edge-of-data uncertainty, and correct burst alignment. Each burst must be triggered before the biomarker occurs, so that the burst's mid-point coincides with it. The core logic is to first calibrate the algorithm on the individual EEG (i.e., stimulate virtually) and then apply calibrated parameters during the stimulation stage to forecast when to trigger a stimulation train. We keep the forecasting short and simple. The algorithm only forecasts the next occurrence of the biomarker based on the calibrated optimal anchor event (preceding theta peak or trough). It uses a calibrated inter-burst interval (IBI) to schedule subsequent bursts within the train. Our total system delay from EEG sampling to TMS coil discharge is 1.7–1.9 ms and processing speed is 0.1–0.4 ms per sample. We partly compensate for edge-of-data uncertainty by adaptive edge removal and calibrated forecasting based on ground truth during a calibration stage.

Fig. 1.

The personalized, state-controlled intermittent theta burst stimulation and electroencephalography (iTBS-EEG) framework (panel A) consists of 1. a high-dynamic range EEG unit with real-time data streaming, 2. a computer unit “PC 2” for EEG monitoring and governing, 3. a computer unit “PC 1” for state control, including parallel software processes for i. data processing and ii. stimulation pulse control, and 4. a TMS unit with 5. real-time neuronavigation and 6. a cobot for coil positioning. Optionally, the EEG unit can receive external event markers via a trigger box, and computer units can receive and send synchronization signals to a cognitive task computer. Panel B depicts protocol schematics, where theta-burst timings, inter-burst interval (IBI), and, by extension, stimulation train length (~ 2 s) are personalized based on each participant's dominant theta rhythm. The experiment starts with a brief EEG recording and a system calibration stage to estimate individual real-time parameters before proceeding to the stimulation stage. Panel C reports the stimulation accuracy for the EEG theta phase-controlled iTBS targeting the left prefrontal cortex across two experiments. The accuracy is shown on the median normalized polar plots as a deviation from the targeted phase, where 0 degrees corresponds to the best accuracy (bars point to the ‘east’).The left column depicts iTBS train initiation accuracy (which depends on theta phase forecasting) and the right column shows the 2nd burst in the train (which depends on the calibrated IBI). Experiment 1 is the in-silico validation in the Dortmund Vital Study dataset (n = 608 participants). The system was calibrated using pre-task resting-state recordings and validated against pre-task (upper row) or post-task recordings (2 hours later; middle row). Experiment 2 is the in-situ validation in our experimental group (n = 10 participants; lower row).

We validated iTBS-EEG in two experiments, targeting theta rhythm (4–8 Hz) phase in the left prefrontal cortex. Experiment 1 virtually simulated phase-controlled stimulation on the Dortmund Vital Study dataset (n = 608, 20–70 years old, [7]), which includes two 3-min resting-state recordings per person made 2 hours apart. We calibrated the algorithm on the first recordings and applied it to both. We report the phasic accuracy of targeting the intended theta phase (Fig. 1C), relative to the ground-truth phase derived post-hoc from zero-phase FIR filtering (stopband 80 dB). For the stimulation train initiation, the median accuracy was 96.2% in the first recording per participant and 95.1% in the second recording. The accuracy of aligning the second burst in a train was 70.2% and 69.0%, respectively. Thus, the algorithm is stable over two hours post-calibration and reliable at each train's start, with no significant differences (all pairwise, uncorrected p-values > 0.2) across age groups (20–35 / 35–50 / 50–70 years old) or sex. See Supplementary Materials for more statistics. Experiment 2 was conducted at Northwestern University (n = 10, 19–39 years old; IRB# STU00223602) with all equipment active but TMS pulses withheld, enabling reliable EEG phase estimation (as done before [4]). Results, reported immediately post-calibration, agreed with Experiment 1: 92.0% median accuracy for train initiation and 70.9% median accuracy for the 2nd burst. Notably, phase accuracy for train start reflects genuine alignment of the first burst to the endogenous theta oscillation. However, for subsequent bursts, the accuracy will likely also depend on the consistency of stimulation-induced entrainment (discussed below), which is outside the scope of the current technical validation. Of note, even if later stimulation bursts in each train keep little phase alignment, delivering all 10 bursts as for iTBS, at minimum, should preserve known benefits.

A practical consideration is that no EEG-controlled method can reliably forecast theta phase for several seconds to maintain mechanistic alignment throughout an iTBS train. We argue that phase-aligning the first few bursts within each train should “seed” the intra-train cascade at optimal excitability. Intracranial recordings show that repetitive bursts produce frequency-specific entrainment persisting briefly past offset [8] and induce dynamic short-term modulation after a few bursts [9]. The subsequent bursts are delivered at the person’s individual theta rhythm, thereby riding the individual dynamics sustained by entrainment from the preceding rhythmic bursts – a “seed-and-sustain” mechanism. Indeed, personalizing iTBS to the individual theta rhythm has been shown beneficial even in open-loop delivery [5]. Single-burst theta-phase-triggered TMS [6] and recent direct hippocampal stimulation [10] further demonstrate that theta-phase-locked stimulation enhances electrophysiological connectivity. Our proposed mechanism combines rationales of personalized rhythm, intra-train entrainment, and phase-locked delivery.

In summary, we show the technical feasibility of EEG state-controlled iTBS over the prefrontal cortex. By aligning each stimulation train's initial bursts to the individual theta rhythm and hypothesizing that the entrainment sustains subsequent bursts, we sidestep the multi-second forecasting problem. Clinical evaluation of this approach is a natural next step.

Highlights.

• Intermittent TBS synchronized to the prefrontal EEG theta rhythm is feasible

• Personalized iTBS-EEG parameters are stable over the typical session time

• Clinical evaluation of iTBS-EEG is a direction for future work