Transcranial ultrasound stimulation of the pulvinar and globus pallidus internus facilitates visual search

Summary

Ultrasound offers a powerful means of modulating human cognition by noninvasively targeting subcortical structures previously accessible only via invasive procedures. While decades of research have mapped cortical circuits of attention, deep hubs such as the basal ganglia and thalamus are increasingly recognized as key nodes in attention networks. We tested whether low-intensity transcranial ultrasound stimulation (TUS) of the globus pallidus internus (GPi) and pulvinar modulates visual search, an attention-dependent task, predicting site-specific effects that reflect distinct basal ganglia-thalamic circuit functions. Focal TUS reduced reaction times, suggesting a facilitation of attention-related task performance. A dissociation emerged across sites: stimulation of both GPi and pulvinar reduced reaction times, but pulvinar yielded more robust benefits for target-present trials at peripheral eccentricities, and improved search efficiency in the same trials. These results support distinct contributions of basal ganglia and thalamic nodes to attention-related behavior and show that TUS can be used to probe circuit-level mechanisms of cognition in humans.

Graphical abstract

Highlights

• •Ultrasound enables causal probing of deep brain circuits with high resolution
• •Stimulation of the globus pallidus internus and pulvinar enhances visual attention
• •Pulvinar stimulation has greater peripheral attention and search efficiency effects

Behavioral neuroscience; Cognitive neuroscience; Techniques in neuroscience

Introduction

Decades of research have mapped the cortical networks underlying attention in humans.^1,2,3,4 However, most human studies of attention have been correlational or based on neurological patients with lesions,^5,6,7,8,9 which limits causal inference about underlying mechanisms. Deep subcortical structures like the basal ganglia and thalamus have been difficult to probe in healthy individuals because they lie beyond the reach of conventional non-invasive methods. As a result, the contributions of these deep hubs to attention remain poorly understood in the intact human brain. Developing methods to causally and focally modulate deep brain activity in humans would fill a critical gap between invasive patient studies and surface-focused neurostimulation, enabling insights into how subcortical circuits actively shape cognition.

Transcranial ultrasound stimulation (TUS) has recently emerged as a technology uniquely suited for non-invasive, focal perturbation of deep brain circuits.¹⁰ Appropriate TUS protocols can induce neural effects that outlast the stimulation period by tens of minutes, permitting the study of “offline” changes in behavior.^11,12 These features position TUS as a means to causally test the roles of deep brain structures in cognition in neurologically normal humans. Early applications in animals and humans have validated that TUS can modulate activity in subcortical targets such as the basal ganglia and thalamus,^13,14 supporting its potential to serve as a non-invasive analog of deep brain stimulation (DBS).

The thalamus, long regarded as a passive relay for sensory information, is increasingly recognized as an active participant in cognitive processes.¹⁵ Thalamic nuclei have extensive reciprocal connections with cortex, placing them in a strategic position to regulate information flow across distributed networks.¹⁶ In particular, the pulvinar nucleus of the thalamus has been implicated in visual attention.^7,9,17 The pulvinar is well-situated to influence attentional orienting and filtering of visual input, given its dense interconnections with visual cortical areas.¹⁸ Consistent with this anatomy, a growing body of work in both non-human primates and humans links pulvinar activity to selective attention.^19,20,21,22 For example, neural recordings and perturbations in monkeys have shown that pulvinar synchronizes cortical activity and can bias visual processing during attention tasks.²³ Inactivation of the pulvinar in monkeys impairs target detection,²⁴ whereas driving pulvinar activity can enhance attention and performance.²⁵ Together, these studies suggest that the thalamus—and the pulvinar in particular—plays a key role in coordinating spatial attention, rather than serving as a mere relay.

By contrast, the globus pallidus internus (GPi), the main basal ganglia output nucleus, has classically been associated with motor functions²⁶ (for instance, GPi is a primary surgical target for movement disorders^27,28). Its potential contributions to cognition have received comparatively limited attention. Anatomical and circuit-tracing studies demonstrate that the basal ganglia are not exclusively motor: they send projections to the prefrontal cortex and other non-motor regions.²⁹ Indeed, the primate GPi is a node in fronto-striatal loops that could influence decision-making and cognitive control.^30,31 Neurophysiological observations in clinical contexts also hint at a cognitive role for GPi—for example, GPi neurons modulate their activity during oddball attention tasks in Parkinson patients.³² Yet, direct evidence for GPi involvement in attentional processes has remained elusive, in part because non-invasive methods to selectively perturb the GPi have not been available. The advent of TUS now enables tests of whether this deep motor structure also contributes to higher order cognitive operations.

Against this backdrop, we leveraged low-intensity TUS to modulate the pulvinar and the GPi in healthy human participants performing a visual search. By targeting these subcortical structures, our aim was to determine how stimulating thalamic versus basal ganglia nodes alters top-down attention-related visual search behavior.

Results

Each participant completed a visual search task (Figure 1A) in three blocks: baseline (Pre), shortly after the first stimulation (Post-1), and shortly after the second stimulation on the same day (Post-2), with TUS targeting either the GPi or the pulvinar bilaterally, with the order randomized across participants, in two sessions conducted at least one week apart. We applied a theta-burst TUS protocol, which has been shown to produce long-lasting effects of up to 1 h after stimulation cessation. To ensure that the spatial distribution of visual stimuli did not influence the potential impacts of TUS across sessions, for each subject (n = 15), a 2D histogram of stimulus positions was computed separately for trials with targets (Figure 1B) and distractors (Figure 1C) during each session (Pre, Post-1, Post-2). Gaze-reported or click-reported screen coordinates were binned into a 10 × 10 grid, using consistent bin edges derived from the full dataset. Each subject’s heatmap was normalized to its maximum bin value to account for individual differences in the number and distribution of trials. Group-level maps represent the average of normalized heatmaps across subjects. Warmer colors indicate bins with higher relative frequency. To test for session-related changes in spatial deployment, paired t tests were performed at each bin between sessions, corrected using false discovery rate (FDR) at α = 0.05. No significant differences were observed across any pairwise session comparison for either targets or distractors, indicating that the spatial distribution of visual stimuli remained consistent across time points.

Figure 1.

Effects of GPi and pulvinar TUS on visual search performance

(A) Experimental paradigm. Participants (n = 15) performed a “T among Ls” visual search task before and after bilateral theta-burst TUS (tbTUS) targeting either GPi or pulvinar. Each session included a baseline block (Pre), a block immediately after the first sonication (Post-1), and a block after the second sonication (Post-2). Set size (1, 4, 8, 16, or 32 items) and target eccentricity (<8°, 8°–30°, >30°) were manipulated.

(B and C) Group-level spatial heatmaps of target and distractor positions across sessions. No significant differences were observed across any pairwise session comparison for either targets or distractors, indicating that the spatial distribution of visual stimuli remained consistent across time points.

(D) Individualized acoustic simulations confirmed focal sonication of GPi and pulvinar at target intensities in the low-intensity neuromodulatory range (color scale = spatial-peak pulse-average intensity (ISPPA), W/cm²). Across subjects, modeled ISPPA was 2–8 W/cm². See also Figure S1.

(E) Target detection accuracy (mean ± SEM) for pulvinar (purple) and GPi (green) remained high across sessions (GPi: Pre = 0.962 ± 0.026, Post-1 = 0.950 ± 0.037, Post-2 = 0.948 ± 0.038; pulvinar: Pre = 0.952 ± 0.035, Post-1 = 0.939 ± 0.054, Post-2 = 0.940 ± 0.041). No within- or between-site differences were significant after FDR correction (all p > 0.5, paired t test). Bars and shaded error margins indicate SEM.

Individualized acoustic simulations, confirming focal sonication of the intended deep target with in-brain intensities in the safe low-intensity neuromodulatory range, are depicted in Figure 1D. More detailed descriptions of the acoustic simulation outcomes for individual subjects are provided in Figure S1, and in our previous work,¹³ in which the same cohort of participants underwent sonication of the same targets, with cognitive performance assessed before and after stimulation using a different paradigm (see also STAR Methods for additional details on the task, TUS protocol, acoustic modeling, safety, and data analysis).

Task accuracy remained high across all sessions and was not affected by stimulation (Figure 1E), indicating that TUS did not affect the probability of finding the target. Reaction time (RT) analyses, however, revealed a richer pattern.

RT distributions were reliably shifted towards shorter RTs following stimulation (Figures 2A–2D). This is most directly visible in the raw cumulative distributions (Figure 3A for GPi; Figure 3B for pulvinar), where the Post-1 and Post-2 curves are systematically shifted leftward relative to baseline, indicating faster responses. In Figures 2A–2D, we plot the subtraction of these curves (Post – Pre), which makes the size and timing of the differences clearer. Notably, these curves diverged most strongly at higher set sizes, consistent with greater modulation under increased task demands. Importantly, a similar pattern was observed when target-present and target-absent trials were analyzed separately, as shown in Figure 3. To exclude the possibility that the observed RT improvements reflected learning or task familiarity, we compared average RTs from the first and second halves of trials within each block. Across all blocks, in both GPi and pulvinar sessions and at both pre- and post-stimulation, no significant differences were found between early and late trials (all p > 0.05, Wilcoxon signed-rank test). This indicates that practice effects did not drive the speeding of responses, supporting the conclusion that TUS was responsible for the shortened RTs.

Figure 2.

Effects of GPi and pulvinar TUS on reaction time and search efficiency

(A–D) Difference plots of cumulative RT distributions (Post − Pre) demonstrate systematic post-stimulation RT acceleration. Negative deflections indicate shifts of the underlying distributions toward shorter RTs. In GPi sessions, shifts were present at Post-1 (A) and strongest at Post-2 (B), particularly at higher set sizes (e.g., set size 32: ΔMedian = −0.478 s, D = 0.184, p = 2.6 × 10⁻⁷, Kolmogorov-Smirnov test). Pulvinar stimulation also accelerated RTs, with significant changes already at Post-1 (C) and the largest effects at Post-2 (D), especially at high load (set size 32: ΔMedian = −0.468 s, KS D = 0.200, p = 2.7 × 10⁻⁹, Kolmogorov-Smirnov test). The corresponding raw cumulative functions, which more intuitively illustrate the leftward shift in RTs, are shown in Figure 3A (GPi) and Figure 3B (pulvinar), where target-present and target-absent trials are also displayed separately.

(E–G) ΔRT (mean ± SEM) by target eccentricity. GPi stimulation produced significant speeding at intermediate eccentricities (8°–30°) in Post-2 (p = 0.0028, paired t test), while pulvinar stimulation yielded stronger improvements at peripheral targets (>30°) in Post-2 (p = 0.0019, paired t test). A significant between-site difference emerged centrally (<8°) in Post-1 (p = 0.03, paired t test).

(H) Search slope effects (s/item ±SEM). GPi slopes declined from 0.0586 to 0.0490 s/item at Post-2 (Δ = −0.0096 ± 0.0105, p = 0.0065, paired t test), while pulvinar slopes decreased more steeply, from 0.0680 to 0.0497 s/item (Δ = −0.0182 ± 0.0121, p = 8.5 × 10⁻⁵, paired t test), with a significant between-site difference (p = 0.035, paired t test). Significance stars: ∗∗∗p < 0.001, ∗∗p < 0.01, ∗p < 0.05; n.s., not significant (all values FDR-corrected). Bars and shaded error margins indicate SEM.

Figure 3.

Cumulative distribution analyses of reaction times for GPi and pulvinar stimulation

(A) GPi. Empirical cumulative distribution functions (ECDFs) of reaction times across set sizes (1, 4, 8, 16, and 32 items) for all trials combined (top), target-absent trials (middle), and target-present trials (bottom). Solid lines indicate Pre, dashed lines Post-1, and dotted lines Post-2 sessions. Insets report FDR-corrected p values from Kolmogorov-Smirnov (KS) tests for Pre − Post-1, Pre − Post-2, and Post-1 − Post-2 contrasts.

(B) Pulvinar. ECDFs for all trials combined (top), target-absent (middle), and target-present (bottom) conditions. Insets show FDR-corrected Kolmogorov-Smirnov p values for each set size and session contrast.

We next examined how these improvements varied as a function of target eccentricity and trial type. Both GPi and pulvinar stimulation reduced average RTs at intermediate (8°–30°) and large (>30°) target eccentricities, but not for central targets (<8°, Figures 2E–2G). The larger benefits observed at more eccentric locations could reflect either a greater number of attentional shifts in the search for the target or simply that longer baseline RTs provided more opportunity for stimulation effects to accumulate. When trials were pooled, no reliable differences emerged between GPi and pulvinar. However, when trials were separated by trial types (Figures 4A and 4B), target-present trials (Figure 4B) revealed a site-specific difference at peripheral locations (>30°), where pulvinar TUS produced stronger improvements than GPi TUS, suggesting that these areas may play distinct roles in supporting performance when targets appear in the far periphery.

Figure 4.

Eccentricity and search efficiency

(A) ΔRT by eccentricity for target-absent trials. Both GPi and pulvinar produced significant speeding at intermediate eccentricities and peripheral locations.

(B) ΔRT by eccentricity for target-present trials. GPi showed facilitation at 8°–30° (ΔPost-2 p = 0.0078, paired t test), while pulvinar selectively improved reaction times at >30° (ΔPost-2 p = 0.0071, paired t test). A significant GPi-pulvinar difference was observed at <8° in Post-1 (p = 0.03, paired t test).

(C) Target-absent trials (left plot): both GPi and pulvinar showed significant slope reductions. GPi: ΔPost-1 = −0.0111 ± 0.0074 s/item (p = 9.1 × 10⁻⁵, paired t test), ΔPost-2 = −0.0146 ± 0.0133 (p = 8.4 × 10⁻⁴, paired t test). Pulvinar: Δpost-1 = −0.0150 ± 0.0204 (p = 0.013, paired t test), ΔPost-2 = −0.0272 ± 0.0187 (p = 1.2 × 10⁻⁴, paired t test). Target-present trials (right plot): pulvinar produced significant slope reductions at post-2 (Δ = −0.0089 ± 0.0110, p = 0.015, paired t test), whereas GPi slope changes did not reach significance (all p > 0.5, paired t test). Bars and shaded error margins indicate SEM.

Search efficiency analyses further supported this dissociation. Slopes of the RT-set size function declined significantly after stimulation at both sites, reflecting increased processing efficiency per distractor, but the effect was stronger for pulvinar. GPi slopes decreased from 0.0586 s/item at baseline to 0.0490 s/item at Post-2, corresponding to an increase in processing rate from 17.1 to 20.4 items/s (Δ = +3.3 items/s). Pulvinar slopes fell from 0.0680 to 0.0497 s/item, corresponding to an increase from 14.7 to 20.1 items/s (Δ = +5.4 items/s). This larger gain for pulvinar was reflected in a significant between-site difference (p = 0.035) (Figure 2H). However, when trials were separated by trial types (Figure 4C), only TUS of the pulvinar led to increased search efficiency in the target-present trials. In the target-absent trials, however, search efficiency was increased following stimulation of both areas (Figure 4C).

To assess individual variability in response to stimulation, we also quantified the global behavioral benefit for each participant as the reduction in median RT from baseline to the post-stimulation blocks. This analysis revealed that the majority of participants showed RT improvement following stimulation, while a subset of non-responders was also present (Figure 5). In the GPi condition, 11 out of 15 participants (73.3%) exhibited a positive behavioral benefit, while 4 participants (26.7%) were classified as non-responders (Figure 5A). The median improvement was 0.146 s (interquartile range = 0.202 s), and the effect was significant across participants (Wilcoxon signed-rank test, p = 0.009). In the pulvinar condition, 10 out of 15 participants (66.7%) showed improvement, while 5 participants (33.3%) did not (Figure 5C). The median improvement was 0.059 s (interquartile range = 0.310 s), and the effect was significant at the group level (p = 0.042). These results demonstrated that while the majority of participants benefited from stimulation, the magnitude of the effect varied across individuals, and non-responders were present in both targets.

Figure 5.

Inter-individual variability in behavioral benefit and its relationship to acoustic intensity

(A) GPi: subject-level RT benefit, defined as the reduction in median reaction time from baseline (Pre) to the average of post-stimulation blocks (Post-1 and Post-2). Participants are sorted by effect size. Positive values indicate faster responses following stimulation (responders). Individual subject identifiers are shown above each bar. Most participants (11/15) exhibited positive behavioral benefit, while a subset of non-responders was observed.

(B) GPi: relationship between RT benefit and subject-specific ISPPA. Each point represents one participant. No significant correlation was observed (Spearman r = −0.10, p = 0.71).

(C) Pulvinar: subject-level RT benefit, as defined in (A). Most participants (10/15) showed improvement, with a subset of non-responders and greater variability in effect size compared to GPi.

(D) Pulvinar: relationship between behavioral benefit and ISPPA. No significant correlation was observed (Spearman r = −0.38, p = 0.16). Behavioral benefit was computed using correct trials only and pooled across target-present and target-absent conditions.

To assess whether inter-individual differences in delivered acoustic intensity contributed to this variability, we examined the relationship between behavioral benefit and subject-specific ISPPA values. No significant correlation was observed in either condition (GPi: Spearman r = −0.10, p = 0.71; pulvinar: r = −0.38, p = 0.16), suggesting that variability in behavioral effects was not explained by differences in delivered acoustic intensity within the range tested (Figures 5B and 5D).

To further investigate inter-individual variability, we also examined the stability of subject-level behavioral effects across the two post-stimulation blocks. Subject-level behavioral benefits were strongly correlated between Post-1 and Post-2 (pulvinar: r = 0.779, p = 0.001, GPi: r = 0.796, p = 0.0006). These results suggest that the behavioral effects are not random or transient at the individual level, but instead reflect stable inter-individual differences in responsiveness to stimulation.

To determine whether the observed RT improvements reflected a speed-accuracy trade-off, we examined the relationship between subject-level RT benefit and changes in accuracy. RT benefit was defined as the reduction in median RT from baseline (pre) to the average of the two post-stimulation blocks (Post-1 and Post-2), while accuracy change was defined as the corresponding difference in accuracy.

No significant correlation was observed between RT benefit and accuracy change for either GPi (Spearman r = −0.27, p = 0.32) or pulvinar stimulation (r = −0.25, p = 0.36), indicating that participants who exhibited faster responses did not show systematic reductions in accuracy (Figure 6). These findings argue against a speed-accuracy trade-off and suggest that the behavioral effects reflect genuine improvements in performance.

Figure 6.

Speed-accuracy trade-off analysis for GPi and pulvinar stimulation

Scatterplots show the relationship between subject-level reaction time (RT) benefit and accuracy change following stimulation for GPi (left) and pulvinar (right). Each point represents one participant. RT benefit was defined as the reduction in median RT from baseline (Pre) to the average of the two post-stimulation blocks (Post-1 and Post-2), such that positive values indicate faster responses following stimulation. Accuracy change was defined as the difference between mean post-stimulation accuracy (averaged across Post-1 and Post-2) and baseline accuracy, such that positive values indicate improved accuracy.

In this study, although a formal a priori power analysis was not performed, post hoc effect size estimates indicated moderate effects (Cohen’s d = 0.57 for pulvinar; Cohen’s d = 0.46 for GPi), calculated as paired effect sizes on the primary behavioral outcome, defined as the within-subject change in RT (difference between mean post-stimulation RT (averaged across Post-1 and Post-2) and baseline RT).

Discussion

Our findings reveal that the human attentional system can be experimentally tuned at its deep subcortical sites. Within the pallido-thalamic network, both GPi and pulvinar stimulation accelerated RTs at intermediate and peripheral target eccentricities. Importantly, a site-specific dissociation emerged in target-present trials, where pulvinar produced a more efficient search and stronger RT improvements than GPi at peripheral locations, suggesting that these structures may differentially support performance depending on spatial demands.

The pulvinar findings converge with a large body of work showing that this nucleus acts as an active attentional hub and contributes to selective filtering and prioritization of relevant inputs.^19,21,23 Human lesion and imaging studies demonstrate its role in suppressing distractors and amplifying relevant signals,^21,22 and primate physiology reveals that pulvinar neurons synchronize cortical networks to bias perception.²³ Recent causal manipulations further support this view; Boshra et al. showed that inducing burst firing in macaque pulvinar enhanced attentional performance.²⁵ Our results extend these findings to healthy humans, providing causal evidence that augmenting pulvinar function sharpens selective attention, particularly under high load and eccentricity.

Several non-mutually exclusive mechanisms could account for the RT speeding following GPi-TUS. First, tbTUS could inhibit GPi output neurons. Such inhibition would remove tonic pallidal inhibition of the thalamus, disinhibiting thalamo-cortical drive and accelerating commitment. This effect mirrors the effective action of GPi-DBS in Parkinson disease.³³ Second, tbTUS could facilitate GPi output neurons. This would be consistent with our prior finding of increased GPi LFP power.¹³ Such facilitation could produce further response speeding, possibly by regulating urgency, which is consistent with the role of the basal ganglia in regulating decision thresholds. Neurophysiological data indicate that pallidal neurons exert an urgency signal that reduces RT of all responses.³¹ GPi outputs to prefrontal cortex²⁹ can exert similar urgency on cognitive processes. Accordingly, GPi could contribute to attentional performance by lowering global response thresholds, rather than improving spatial filtering. These possibilities can be disambiguated in future experiments combining TUS with simultaneous neurophysiological recordings or pharmacological blockade in animal models.

By directly stimulating these deep structures in healthy participants, we show that attentional control can be modulated non-invasively. This work moves beyond correlational mapping and patient studies, offering a direct demonstration that subcortical attention circuits can be tuned in the intact human brain. Importantly, the cohort of participants in the present study also completed a stop-signal task in our prior work,¹³ enabling within-subject comparisons across tasks. In that study, GPi stimulation impaired response inhibition measured by longer stop-signal RT. In the current study, stimulation reduced RTs in the visual search task. On the other hand, in the previous study pulvinar stimulation did not affect stop-signal RTs,¹³ whereas in the current study pulvinar stimulation significantly reduced RTs in the visual search task.

Opposite effects that vary with task are difficult to explain through nonspecific factors such as drowsiness or general arousal, which would be expected to produce consistent directional changes across tasks. Instead, this dissociation indicates that stimulation interacts with task-specific circuit demands. The dissociation between pulvinar and GPi further supports this interpretation. Further evidence against a global state explanation is the improved behavioral performance even when participants occasionally reported drowsiness after stimulation. Moreover, the absence of significant differences in RTs between the first and second halves of trials within each block suggests that practice effects did not drive the observed improvements. It should be noted that participants were blind to stimulation target. Nevertheless, the absence of a sham condition and lack of blinding to stimulation timing may introduce expectation-related effects, which should be addressed in future studies.

Previous work using theta-burst TUS over motor cortex has shown that its neuromodulatory effects can persist for 30–90 min.¹² Other stimulation protocols have also been shown to induce long-lasting effects in both superficial and deep brain circuits.^14,34 Our design, which sampled behavior at two post-stimulation blocks (post-1 and post-2), allowed us to estimate the temporal evolution of attentional benefits within this window. RT improvements were present at Post-1 and were further amplified at Post-2, suggesting either cumulative effects of repeated stimulation, alignment with a more optimal post-stimulation time window, or both. Since we used a single set of stimulation parameters (500 kHz, ISPPA ≈ 2–8 W/cm² in brain, 120 s per hemisphere), we cannot map the full dose-response curve. Future work should systematically manipulate acoustic intensity, burst duration, and inter-burst spacing to chart the relationship between TUS “dose” and behavioral outcomes. Such studies will be critical to optimize protocols for both basic research and clinical applications.

From a dynamical systems perspective, non-invasive stimulation can influence brain function by modulating both local excitability and large-scale network dynamics. Computational and whole-brain modeling studies have shown that external stimulation can alter the excitatory-inhibitory balance and synaptic gain, leading to both immediate and short-lasting effects, and can shift neural systems between distinct dynamical regimes, including changes in stability and attractor structure.³⁵ Within this framework, the effects observed here likely reflect a combination of transient changes in excitability and short-lived reconfiguration of task-relevant network states. The dissociation between pulvinar and GPi supports circuit-specific modulation of network dynamics rather than a global effect, raises specific questions to be answered in future nonhuman primate studies, and provides useful constraints for future model development.

Methodologically, this study showcases the potential of TUS as a tool for cognitive neuroscience. Its millimeter-scale focality and ability to target structures several centimeters beneath the cortical surface allow manipulations previously restricted to invasive approaches such as DBS. Both TUS and DBS can modulate activity within deep circuits, but they do so via different biophysical routes and potentially different network effects. DBS delivers high-frequency electrical pulses that directly influence membrane potentials and are often interpreted as disrupting or regularizing pathological oscillations in disorders such as Parkinson disease.²⁸ In contrast, low-intensity TUS delivers mechanical energy that is thought to modulate neural activity indirectly via mechanosensitive ion channels and astrocytic signaling.^36,37 For example, a study demonstrated that focused ultrasound can activate astrocytic TRPA1 channels and trigger downstream neuromodulatory cascades,³⁶ and another study showed that ultrasound can excite cortical neurons via mechanosensitive calcium accumulation and amplification of ion channel activity.³⁷ Various neurotransmitter systems have also been implicated in the neuromodulatory effects of TUS.^10,34 These mechanisms provide a plausible substrate for the offline effects observed here, in which brief stimulation of GPi or pulvinar leads to lasting changes in the excitability and coordination of their associated cortical networks.

Computational work on network-level oscillations and neuromodulation^38,39 has begun to assess how DBS can move abnormal brain states toward healthier dynamical regimes. Our results suggest that TUS could offer a non-invasive way to probe and potentially reshape similar circuits, albeit through a different, more “modulatory” mechanism. Understanding how these two interventions converge or diverge at the level of whole-brain dynamics remains an important direction for future modeling and empirical work.

Our results raise the intriguing possibility that deep TUS could be leveraged not only to probe but also to ameliorate attentional deficits in clinical populations. In attention-deficit/hyperactivity disorder (ADHD), evidence points to abnormalities in thalamo-cortical connectivity and impaired filtering of distractors.^21,22 By enhancing pulvinar-mediated spatial filtering, TUS might help reduce distractibility and improve sustained attention.

In Parkinson disease, the GPi is a well-established DBS target for motor symptoms,^27,28 and recent work suggests that theta-burst TUS to basal ganglia circuits can induce plastic changes in patients.⁴⁰ Our finding that GPi TUS in healthy individuals speeds responses without compromising accuracy resonates with its potential role in alleviating bradykinesia or cognitive slowing. Importantly, our study was conducted in young, neurologically healthy adults, with a single-session design. We therefore view our findings as proof-of-principle for non-invasive modulation of deep attentional circuits rather than as directly translatable treatment protocols. Key steps toward translation will include individualized acoustic modeling in patients, multi-session safety and efficacy studies, and careful integration with existing interventions such as medication and DBS.

More broadly, if selective stimulation can enhance attention, it may also be possible to boost memory, mood, or other cognitive and affective domains by targeting their circuits. Once protocols are developed that yield longer-lasting effects, TUS could provide a noninvasive alternative to DBS, with both scientific and therapeutic impact.

Limitations of the study

Several limitations should be considered. The absence of a sham condition and lack of blinding to stimulation timing leave open the possibility of expectation-related effects. In addition, despite individualized acoustic modeling, the spatial resolution of TUS is limited, and stimulation likely extended beyond the intended nuclei; therefore, the results should be interpreted at the level of distributed pallido-thalamic circuits rather than single targets. Finally, the lack of concurrent neural recordings and the single-session design in a modest sample limit mechanistic interpretation and generalizability.

Resource availability

Lead contact

Further information and requests for resources should be directed to and will be fulfilled upon reasonable request by the lead contact, Hamidreza Ramezanpour (hamidram@yorku.ca).

Materials availability

This study did not generate any novel reagents.

Data and code availability

• •All data reported in this paper are included in the published article and its supplemental information files. The acoustic simulation data are publicly available at https://doi.org/10.6084/m9.figshare.31904413.
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

The study was funded by Canadian Institutes of Health Research grants (FDN 154292, PJT 198046) and a Natural Sciences and Engineering Research Council of Canada grant (RGPIN-2020-04176) to R.C.

Author contributions

Conceptualization, H.R., G.D., and R.C.; methodology, H.R., G.D., R.A., C.S., J.-F.N., S.P., J.D.S., A.M.L., and R.C.; investigation, H.R., G.D., R.A., C.S., J.-F.N., and S.P.; formal analysis, H.R.; software, H.R., G.D., R.A., and S.P.; validation, H.R., G.D., R.A., J.D.S., A.M.L., and R.C.; data curation, H.R., G.D., R.A., and C.S.; visualization, H.R. and G.D.; resources, H.R., S.P., J.D.S., A.M.L., and R.C.; project administration, H.R., G.D., R.A., and R.C.; writing – original draft, H.R.; writing – review and editing, H.R., G.D., R.A., C.S., J.-F.N., S.P., J.D.S., A.M.L., and R.C.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Software and algorithms
MATLAB	Mathworks	https://www.mathworks.com/products/matlab.html
BabelBrain	BabelBrain	https://proteusmrighifu.github.io/BabelBrain/
Deposited data
Acoustic simulation data	This paper	https://doi.org/10.6084/m9.figshare.31904413
Other
NeuroFUS Pro ultrasound system	Brainbox	https://brainbox-neuro.com/products/neurofus

Experimental model and study participant details

Participants

Fifteen healthy adult volunteers (age 18–35 years; 4 male, 11 female) participated in the study. All were neurologically healthy, and provided written informed consent. They all had normal or corrected-to-normal vision. Vision correction methods (e.g., glasses or contact lenses) were permitted and consistent across sessions. Fourteen of the fifteen participants were right-handed. The University Health Network Research Ethics Board approved the study in accordance with the Declaration of Helsinki. The cohort of participants in the present study also completed a stop-signal task in our prior work,¹³ enabling us to use identical stimulation and targeting procedures, and to make within-subject comparisons across tasks. The sample size was determined based on prior transcranial ultrasound stimulation studies in humans, including our previous work using a similar within-subject design,¹³ which demonstrated measurable behavioral effects with comparable cohort sizes.

Method details

Ultrasound stimulation and acoustic modeling

Stimulation was delivered with a single-element, 500 kHz focused transducer (Sonic Concepts, Bothell, WA) coupled to the scalp via ultrasound gel. Each session targeted one structure bilaterally, the internal segment of the globus pallidus (GPi) or the pulvinar nucleus of the thalamus, using neuronavigation (Brainsight, Rogue Research, Montreal, Canada). The order was randomized across participants, with the two sessions conducted at least one week apart. A theta-burst transcranial ultrasound stimulation (tbTUS) protocol was applied, consisting of 20 ms bursts at 50 Hz repeated every 200 ms (10% duty cycle) for 120 s per hemisphere. The spatial peak pulse average intensity in water was 30 W/cm².

Target locations were identified from individual T1- and T2-weighted MRI scans, co-registered to a standard atlas (AAL3) and refined to each participant’s anatomy. For GPi, average MNI coordinates were ±20, −8, −4; for pulvinar, ±11, −33, 6. Coordinates were transformed into subject space and optimized for skull geometry using individualized acoustic simulations in BabelBrain.⁴¹ Finite-difference time-domain methods incorporating skull density and cortical surfaces were used to estimate in-brain intensities and focal profiles. Across participants, modeled focal spots were ∼24 mm axially and 4–5 mm laterally at −6 dB, with estimated in-brain ISPPA between 2 and 8 W/cm². However, intensities above the conservative thermal safety threshold extended several millimeters into surrounding structures including GPe and adjacent thalamic nuclei. Thus, our protocol should be understood as selectively biasing activity within a deep subcortical circuit rather than isolating a single nucleus.

Acoustic simulations predicted minimal thermal effects, with maximum rises of 0.16 °C in the brain, 0.25 °C in the skin, and 0.30 °C in the skull. Transducer adjustments (Δx, Δy, Δz) were required to compensate for skull-induced distortions, reducing the distance between intended and effective targets. Within the target region, ISPPA values ranged from 2.5 to 7 W/cm² and ISPTA from 0.25 to 0.7 W/cm², with mechanical indices between 0.38 and 0.67. All parameters remained well within FDA and ITRUSST safety limits.⁴²

Visual search task

The behavioral paradigm was a classic “T among Ls” visual search task. On each trial, a rotated “T” target was either present or absent among distractors. Set size was varied randomly across 1, 4, 8, 16, or 32 items. Target positions were randomized across eccentricities: central (<8°), intermediate (8–30°), and peripheral (>30°). Each trial began with a small central fixation cross that participants were instructed to hold for 500 ms, after which the target and distractors were displayed. Participants sat approximately 70 cm from the monitor and indicated, as quickly and accurately as possible, whether the target was present or absent using two different response keys on a keyboard. Trials were separated by a 1-s inter-trial interval, and responses were accepted within a maximum of 10 s. Each experimental session consisted of three blocks: Pre (baseline), Post-1 (immediately following sonication of the first hemisphere), and Post-2 (following sonication of the contralateral hemisphere). Participants were not blinded to the timing of stimulation, as the transducer was removed after stimulation, making the end of stimulation apparent. However, participants were not informed about the stimulation target in each session (e.g., GPi vs. pulvinar) and were therefore blind to the specific experimental condition. The behavioral task was performed after the stimulation period to assess after-effects of TUS. Each block contained ∼300 trials, balanced for target presence/absence and eccentricity. Although tbTUS effects typically emerge within 5 min and peak around 10–15 min post-exposure, the length of each block ensured overlap with this window. In addition, the second post-sonication block was conducted immediately after a second round of tbTUS, ensuring that stimulation effects were still present and not washed out, consistent with prior studies showing effects lasting up to 1 h post-stimulation.^11,12,13

Quantification and statistical analysis

Analyses were restricted to correct trials. Reaction times (RTs) were characterized with empirical cumulative distribution functions (ECDFs) at each set size. Pre vs. Post-1, Pre vs. Post-2, and Post-1 vs. Post-2 comparisons were performed using two-sample Kolmogorov–Smirnov (KS) tests. Distributional shifts were further quantified as differences in median RT (ΔMedian).

Eccentricity effects were analyzed by computing ΔRT (Post–Pre) separately for central, intermediate, and peripheral bins. Within-site effects were tested against zero (one-sample t-tests) and between-site differences were assessed with paired-sample t-tests.

Search efficiency was estimated by regressing RTs against set size. The intercept reflected general response speed independent of load, while the slope indexed efficiency (s/item). ΔSlopes were tested within site and contrasted between sites. All p-values were corrected for multiple comparisons using the Benjamini-Hochberg FDR (q = 0.05).