Transcranial ultrasound stimulation of the right globus pallidus interna in schizophrenia: A randomized, sham-controlled pilot study

Abstract

Functional abnormalities of the globus pallidus interna (GPi) are linked to psychosis severity and treatment response in schizophrenia. Transcranial ultrasound stimulation (TUS) is a noninvasive method for precisely targeting deep brain structures. The effects of GPi-targeted TUS on psychosis had not been examined. We conducted a random-order, sham-controlled, crossover pilot study assessing safety, tolerability, and preliminary efficacy of right GPi TUS for psychotic symptoms. Ten outpatients (mean age 37.5; 50% female) with schizophrenia or schizoaffective disorder and persistent psychosis received one TUS and one sham session, one week apart. TUS was MRI-guided, targeting right GPi with a 350 kHz transducer (100 Hz pulse repetition frequency, 5% duty cycle, in situ I_spta 400 mW/cm²). Outcomes—delusion experience sampling assessment (DESA) and auditory hallucinations rating scale (AHRS)—were measured before, during, and immediately, 1 h, 1, and 7 days after stimulation. DESA scores decreased significantly more after TUS than sham (p = 0.002, Cohen’s d = 0.68); AHRS and cognition changes were not significant. Mild transient adverse effects occurred in both conditions; none were serious. GPi-targeted TUS was safe, well-tolerated, and associated with a preliminary reduction in delusional severity, warranting further study in larger, randomized trials. ClinicalTrials.gov: NCT05643196, registered 21 November 2022.

Introduction

Neuroimaging and electrophysiological studies have implicated the cortico-striatal-pallido-thalamic (CSPT) circuit in schizophrenia¹. A key component of the CSPT circuit is the globus pallidus interna, which is the primary output structure of the dorsal striatum and conveys the dopaminergic indirect pathway linking the striatum with the thalamus. Our group studied unmedicated individuals with first-episode schizophrenia and found right anterior GPi functional hyperconnectivity compared to controls². The right anterior GPi connectivity to the dorsal anterior cingulate gyrus was positively associated with severity of psychotic symptoms and strongly predicted treatment response after 8 weeks², supporting the right GPi as a target for neuromodulation.

Low-intensity transcranial ultrasound stimulation (TUS) is a potentially ideal method to probe the CSPT circuit, since TUS can target deep structures with high resolution. TUS modulates neuronal excitability in animals^3–5. In humans, TUS applied to both cortical and subcortical structures has demonstrated neuromodulatory effects without serious adverse effects^6–10. Cain et al. ¹¹ sonicated the left GP in healthy humans and found a significant reduction in BOLD signal in the GP, thalamus, and connected structures during stimulation. To our knowledge, the only published report of TUS in psychosis targeted the nucleus accumbens and caudate head in three individuals with schizophrenia and found a significant reduction in hallucinations and a decrease in striatal–superior temporal cortex functional connectivity immediately following sonication¹². The only other published report of TUS in schizophrenia targeted the DLPFC and focused on negative symptoms¹³. To examine the effects of right GPi-targeted TUS on psychotic symptoms, we conducted a random-order, sham-controlled crossover pilot study.

Results

TUS targeting and characterization

Participants were blindly randomized and underwent one active TUS and one sham session in random order, separated by one week. During the active sonication sessions, MRI-based neuronavigation was used to target the right GPi with pulsed TUS for 10 min using a custom-made single-element fixed-depth 350 kHz transducer (see the “Methods” section), as previously described¹⁴. The ultrasound pulse was transmitted in 0.5 ms rectangular pulses. The pulse repetition frequency (PRF) was set to 100 Hz, corresponding to a pulse repetition interval of 10 ms and a 5% duty cycle. The 10-min sonication protocol consisted of ten cycles of 30-s-on and 30-s-off sonication.

Prior to clinical application, we first empirically validated the system’s acoustic output in free water to evaluate focal dimensions, targeting depth and skull attenuation (Fig. 1). We extracted from the empirical free field mapping dataset the axial position of the spatial peak pressure (P_sp) located at 64.9 ± 0.5 mm from the transducer’s exit plane, a −3 dB focal volume of 27.5 ± 0.5 mm axially and 5.6 ± 0.2 mm laterally (expanding to 38.5 ± 0.5 and 7.7 ± 0.2 mm, respectively, at −6 dB). At the output level, the system produced a P_sp of 1.34 MPa, corresponding to a free water spatial peak pulse averaged intensity (Isppa) of 60.1 W/cm². A total of eight empirical experiments across two human skull specimens captured the ultrasound pressure transmission through the temporal suture. Each experiment was an independent comparative assessment of the pressure transmission through the skull, consisting of a baseline acoustic pressure map measured in free water, followed by measurements after the introduction of the skull. The pressure attenuation range calculated for the 350 kHz frequency was 59–65%, corresponding with an intensity attenuation of 83–88% (Fig. 1d) and an estimated in situ spatial peak temporal averaged intensity (I_spta) range of 0.29–0.4 W/ cm². The temporal suture exhibits a relatively large radius of curvature, which may reduce refraction for orthogonal beams. The acoustic pressure field at the focal plane with and without the skull yields an average beam displacement of 1 mm (±0.8 mm).

Fig. 1. TUS calibration in free water and through human skull.

a A projection of the transducer focus on a T1 MRI scan (axial view). b Axial profiles of the normalized pressure distribution along the axial and radial directions. The dashed and dotted lines show the location of the beam maximum and the −3 and −6 dB focal dimensions. c A normalized pseudocolor plot of the plano-concave, Ø85 mm aperture lens’ output, sampled in the XZ plane. The empirical acoustic field was measured over a 30 × 120 mm domain with a spatial resolution of 12 points per wavelength. d A comparison of two consecutive scans of the same XY plane. The pressure transmission factor in each experiment was calculated relative to the free-water scan prior to skull insertion.

Baseline characteristics

The participants (N = 10) had a mean age of 37.5 years (SD = 9.8). Fifty percent were male. Four participants identified as African American, three as Asian, one as Caucasian, one as mixed, and one as other. Their mean baseline BPRS total score was 64.4 (SD = 8.7), and the positive symptoms subscale score was 21.8 (SD = 4.7). Six participants were taking antipsychotic medication (paliperidone, aripiprazole, quetiapine, haloperidol) and remained on the same dose throughout the study; four (40%) were not taking antipsychotic medication at the time of the study.

Clinical outcomes

Total DESA scores decreased after TUS but not sham (mean change, SD: −1.2, 3.2 after TUS; 0.4, 3.1 after sham; Fig. 2). In the mixed-effects model including all post-baseline observations and adjusting for treatment, time, and session order, TUS was associated with a significantly greater reduction in total DESA score compared with sham (estimate = −1.56, 95% CI: −2.54 to −0.58; p = 0.002; Cohen’s d = 0.68; Table 1). No significant treatment effect was observed for AHRS–HCS scores (estimate = −0.40, 95% CI: −1.18 to 0.38; p = 0.31; Table 1). The change from baseline in the Montreal Cognitive Assessment (MoCA) scores did not differ significantly between treatments (mean change, SD: 0.2, 2.1 post-TUS and −1.0, 2.4 post-sham; p = 0.06).

Fig. 2. Mean change in total DESA score from baseline.

Huber mean change in DESA scores from baseline across stimulation time points (during, immediately after, one-hour after, one-day after, and seven-days after sonication) in the sham and TUS groups. Negative values indicate symptom improvement.

Table 1.

Mixed-effects model of AHRS–HCS and DESA scores

	Change in AHRS–HCS since baseline			Change in total DESA score since baseline
Covariate	Estimate	95% CI	p-value	Estimate	95% CI	p-value
(Intercept)	−0.76	[−2.35, 0.83]	0.346	1.15	[−1.16, 3.47]	0.325
Treatment	−0.40	[−1.18, 0.38]	0.312	−1.56	[−2.54, −0.58]	0.002
Time	0.06	[−0.22, 0.34]	0.667	−0.02	[−0.37, 0.32]	0.887
Session order	−0.12	[−0.90, 0.66]	0.761	−0.48	[−1.46, 0.50]	0.334

Estimated coefficients, 95% confidence intervals (CI) and p-values from a mixed-effects model with Gaussian error, controlling for treatment, visit and period, and a random effect for participant ID.

DESA delusion experience sampling assessment, AHRS–HCS auditory hallucinations rating scale–hallucination change score.

The exploratory interaction model estimated the difference between TUS and sham in DESA change from baseline at each post-stimulation time point. The estimated TUS–sham difference was −1.90 at 1 day post-stimulation (95% CI: −4.09 to 0.29; nominal p = 0.09; Holm-adjusted p = 0.36; Table 2) and −2.70 at 7 days post-stimulation (95% CI: −4.89 to −0.51; nominal p = 0.016; Holm-adjusted p = 0.08; Table 2). These exploratory results did not suggest a clear diminution of the estimated effect over the 7-day post-stimulation observation period. The largest estimated effect occurred at day 7, although this exploratory time-specific effect did not remain statistically significant after Holm adjustment.

Table 2.

Model-based mean difference from treatment-by-time mixed-effects model

Time point	Mean difference (TUS-Sham)	SE	95% CI	Nominal p-value	Holm-adjusted p-value
D	−1.00	1.12	[−3.19, 1.19]	0.370	0.741
P	−1.80	1.12	[−3.99, 0.39]	0.107	0.355
P1H	−0.40	1.12	[−2.59, 1.79]	0.720	0.741
P1D	−1.90	1.12	[−4.09, 0.29]	0.089	0.355
P7D	−2.70	1.12	[−4.89, −0.51]	0.016	0.078

Mean difference is TUS minus Sham. Estimates come from a treatment-by-time mixed-effects model adjusted for IV period and including a subject-level random intercept. Standard errors are model-based. They are identical across timepoints because the fitted mixed-effects model assumes a common residual variance and the design is balanced, with 10 Sham and 10 TUS observations at each timepoint. Raw standard deviations may differ by timepoint. Holm-adjusted p-values are adjusted across the five time-specific TUS vs. Sham contrasts.

Bold values refers to the time point that reached significance.

Randomization and sham effect

After study completion, six participants (60%) correctly identified the stimulation condition, which was not statistically distinguishable from chance (p = 0.75). We conducted an exploratory sensitivity analysis evaluating whether correct participant guesses were associated with DESA change. Correct guessing was not associated with DESA change after adjustment for treatment, assessment time, and intervention period (estimate = −0.13, 95% CI −2.41 to 2.15), and there was no evidence that the TUS versus sham difference varied by guess correctness (interaction estimate = 0.45, 95% CI −1.70 to 2.61).

Descriptively, there was no clear evidence of systematic order/carryover. In the TUS→sham sequence, mean DESA at the second-session baseline was slightly lower than at the first-session baseline (15.0 vs. 16.4; mean difference −1.4, SD 2.8; 95% CI ≈ −4.9 to 2.1). In the sham → TUS sequence, session-baseline DESA scores were similar across periods (12.4 vs. 12.6; mean difference −0.2, SD 4.8, 95% CI ≈ −6.1 to 5.7). Given the small sample size, these order/carryover summaries were interpreted descriptively rather than tested formally.

At the individual level, trajectories were heterogeneous, with no consistent pattern that would suggest that participants improved simply as a function of study period or that sham responses uniformly reflected persistent improvement after prior TUS (Fig. 3).

Fig. 3. Individual-level total DESA scores by randomized treatment sequence.

Each participant’s raw DESA scores across both crossover periods, with the dotted vertical line separating session 1 from session 2 and point color indicating TUS or sham. At the individual level, trajectories were heterogeneous, with no consistent pattern suggesting that all participants improved simply as a function of study period or that sham responses uniformly reflected persistent improvement after prior TUS.

Adverse events

No serious adverse event occurred. Seven participants after TUS and all 10 after sham reported an increase in at least one item on the systematic assessment for treatment emergent side effects (SAFTEE) scale, including fatigue, headache, gastrointestinal discomfort, and increased delusions or hallucinations (Table 3).

Table 3.

Adverse events measured by SAFTEE

Symptom	Severity	Post-TUS	Post-sham
Increase in delusions	Mild	2	2
Increase in hallucinations	Mild	1	2
Gastric-intestinal discomfort	Mild	1	2
Fatigue	Mild	2	1
Headache	Mild	1	2
Nervousness	Mild	1	0
Insomnia	Mild	0	1
Dizziness	Mild	0	1
Sensory abnormality	Mild	0	1
Derealization	Moderate	0	1
Impaired coordination	Mild	0	1
Muscle cramps	Mild	0	1

Values represent the number of participants reporting each adverse event. Severity: mild = 1-point increase on SAFTEE; moderate = 2-point increase on SAFTEE.

Discussion

In this pilot study, TUS was well tolerated and produced a signal consistent with a possible treatment effect on delusions in patients with chronic psychosis.

The decrease in delusions may reflect global inhibitory modulation of the CSPT circuit via TUS targeting of the right GPi. This aligns with fMRI findings in healthy subjects in which the GP was sonicated using similar stimulation parameters¹¹. TUS may regulate the balance of inhibition and excitation in stimulated areas through mechanisms that have yet to be fully clarified, such as direct effects on the membrane or activation of mechanosensitive calcium- and voltage-gated ion channels^15,16. Its effects are cell-type and dose-dependent^7,17, and may produce persistent neuroplastic changes¹⁷. Consistent with this, our exploratory analysis found a nominally significant reduction in delusion severity at day 7 post-stimulation. Although replication in a larger, confirmatory trial is warranted, precedent for durable TUS effects exists: repeated TUS stimulations (6 sessions over 2 weeks) in patients with depression have produced persistent symptom relief for up to 4 weeks¹⁸.

We observed a reduction in delusion severity over time, with a maximum reduction at day 7. Although TUS is known to produce immediate neurobiological effects, there is emerging literature on long-lasting changes after TUS. These “offline” changes can persist for hours after stimulation and may reflect LTP/LTD-like plasticity mediated by AMPA and NMDA glutamatergic receptor modulation and postsynaptic Ca²⁺-dependent changes in receptor properties^19,20. Durable therapeutic effects of TUS on deep brain circuits have also been observed in clinical contexts: a single session of TUS targeting the anterior cingulate cortex produced pain relief persisting through day 7²¹, and deep brain ultrasonic neuromodulation of the subcallosal cingulate in treatment-resistant depression induced remission lasting up to 44 days²², with neuroplastic circuit reset proposed as the underlying mechanism in both cases. We propose that an analogous process underlies the delayed peak observed here: TUS-induced plasticity within the GPi and its downstream cortico-striatal-pallido-thalamo-cortical circuitry may have continued to consolidate after stimulation, reaching a functional threshold sufficient to reduce aberrant salience—and with it, delusion severity—by day 7.

Only mild adverse effects occurred, distributed evenly between TUS and sham, consistent with prior human studies. Legon et al. ⁶ surveyed participants in multiple ultrasound neuromodulation studies and found that none experienced serious adverse effects; 7 of 64 participants experienced mild to moderate symptoms potentially related to TUS, including transient neck pain, anxiety, muscle twitches, and attention problems—adverse effects similar to those reported with other forms of non-invasive brain stimulation such as transcranial magnetic stimulation. There was no significant effect of TUS on cognition as measured by the MoCA. In prior studies, cognitive improvement has been reported with sonication of the substantia nigra, mesial and ventromedial anterior temporal lobe, frontoparietal cortex, and hippocampus in humans^10,23.

There are several limitations. First, due to technical issues, the fMRI data were not interpretable. Without neuroimaging confirmation of target engagement, we cannot conclude that the GPi was modulated, which limits the mechanistic interpretation of the findings. Second, DESA was designed for ecological momentary assessment of delusional severity over brief intervals and has not been validated as a measure of clinically significant treatment response. Accordingly, the clinical meaningfulness of the observed change remains uncertain, even though the estimated between-condition effect size was moderate (Cohen’s d = 0.68) and comparable to the efficacy of antipsychotic drugs versus placebo (effect size = 0.47)²⁴. Third, blinding was assessed only with a binary forced-choice questionnaire, which limits the interpretation of blinding integrity. Six out of the 10 participants correctly guessed the condition assignment. Although this result was not statistically distinguishable from chance in this small sample, expectancy effects cannot be excluded, and the study should not be considered fully double-blinded. Fourth, we evaluated only a single TUS session over 7 days, and brief clinical scales were selected to allow repeated within-session assessments; repeated sessions with more extensive assessments are needed to fully evaluate the extent and time course of TUS effects. Fifth, the stimulation dose was not individualized on the basis of skull aberration and attenuation, and individualized parameter adjustment may improve targeting accuracy and dose optimization. Finally, this was a small pilot study with only 10 participants. The findings should be considered preliminary and require replication and extension in larger randomized controlled trials.

Methods

Study design and assessments

The study was approved by the New York University Grossman School of Medicine Institutional Review Board and registered at ClinicalTrials.gov (NCT05643196; full protocol and analysis plan available at https://clinicaltrials.gov/study/NCT05643196), registered 21 November 2022. Please see Supplement for CONSORT diagram (Fig. S1) and checklist (Table S1). The study was conducted in accordance with the Declaration of Helsinki. Participants completed the written informed consent process with a study psychiatrist. Capacity to consent was evaluated by the psychiatrist in addition to scoring 100% correct answers on a 10-item quiz about the study procedure, risks, and alternatives to participation. Eligible participants were enrolled and assigned a simple randomization code by a statistician. After a baseline MRI scan to guide neuronavigation, participants underwent one TUS and one sham session in random order, separated by one week. The bioengineer used MATLAB to generate a randomization code for assigning the order of TUS and sham sessions. All other study team members and participants were blinded until completion of the study.

Participants

Individuals aged 18–55 with schizophrenia or schizoaffective disorder were recruited by clinician referral and online advertising between May 1, 2023 and Jan 31, 2024. Eligible participants scored 4 (moderate) or greater on at least one psychosis item of the brief psychiatric rating scale (BPRS) in the absence of substance use disorders, major mood disorders, suicidal ideation, pregnancy, significant medical or neurological illness, or contraindication to MRI or sonication. Participants were required to be taking a stable dose of medication or be medication-free for at least 4 weeks prior to enrollment. They continued their pre-enrollment medication treatment and stayed medication-free throughout the study. A total of 15 participants were screened and 11 were enrolled. One participant withdrew after the baseline MRI scan and before sonication. Ten participants completed all study procedures and were included in analyses.

The AHRS²⁵ and DESA²⁶ were completed before, during, immediately after, one hour after, one day after, and seven days after each session. The MoCA was completed at baseline and 1 h after each session. SAFTEE and Columbia Suicide Severity Rating Scale (CSSRS) were completed before and after each session, as well as 1 day and 1 week after sessions. Resting-state fMRI data were collected before and after each session; however, due to technical issues, we were unable to analyze the fMRI data and report only the behavioral results.

TUS system and output validation

The TUS system was developed for targeted GPi sonication, comprising a single-element fixed-depth transducer daisy-chained through a signal generator, RF power amplifier, and bidirectional coupler for output monitoring. The piezoelectric crystal (SMD85T6R111, 350 kHz, STEMiNC, Davenport, FL, USA) was bonded to a plano-concave acoustic lens, which was machined from Delrin to provide a 44 mm spherical radius and an 85 mm aperture diameter. The spherical lens’ apex is recessed at 32.8 mm in the axial direction from the transducer’s exit plane, with a total lens height of 37.9 mm.

The driving system was matched to 50 Ω (VSWR ≤ 2:1, rated by the RF power amplifier) and is composed of a wave generator (SDG 1032X, Siglent Technologies NA, Inc., OH, USA) and a broadband 200 W power amplifier (1020L, Electronics & Innovation, Ltd., Rochester, NY, USA).

An MRI-driven stereotactical neuronavigation apparatus superimposed a real-time spatial link between the transducer and the patient in the treatment room. The patient’s T1-weighted structural MRI scan is virtually represented using 3D Slicer²⁷. The patient’s MRI data was registered into the software using MRI-visible fiducial markers. Reflective spherical markers are anchored on the transducer and the subject’s head in a pattern that ensures their visibility to an optical tracking camera system (NDI Polaris Vicra, Northern Digital, Inc., Waterloo, ON, Canada). Segmentation slices of the sagittal, frontal, and transverse planes are used to identify the transducer’s spatial position and orientation, which is monitored in real time.

Head immobilization of a sitting patient was achieved by affixing multiple Velcro-stitched Styrofoam blocks around their head. The fixed focal-depth transducer was then positioned with a 6-degree-of-freedom lockable articulating arm while targeting of the right GPi was closely monitored by the operator. Upon reaching the desired spatial position and orientation of the transducer, the articulating arm was locked, and the transducer was retracted 5 cm away from the scalp. The transducer cavity was covered with a latex membrane and filled with degassed, deionized water to mediate ultrasound transmission. The patient’s hair was gently combed with ultrasound gel to eliminate any air bubbles, and the transducer was then repositioned to the sonication site. Prior to sonication, the transducer orientation was adjusted with another independent, precise ball-joint positioning mechanism. The correct targeting was closely monitored by the operator prior to each sonication cycle. Water bag integrity was maintained for the course of the treatment (~15 min). Sham stimulation was delivered by creating an air layer in the water balloon between the transducer and the skull, and operating the TUS system with the same parameters.

The free-field acoustic parameters were empirically measured in a 760 × 305 × 470 mm glass water tank lined with an acoustic-absorbing Neoprene rubber sheet (McMaster-Carr, Elmhurst, IL, USA). A capsule hydrophone (HGL-0085, Onda Corporation, Sunnyvale, CA, USA) and a submersible preamplifier (AG2010, Onda Corporation, Sunnyvale, CA, USA) were calibrated by the manufacturer shortly before conducting the measurements. The hydrophone was suspended and maneuvered by a linear 3-axis positioning system (Velmex Inc., Bloomfield, NY, USA). A pre-scan alignment procedure ensured that (a) the hydrophone’s axis was perpendicular to the transducer’s exit plane and (b) that the scanned x–z plane intersected the beam axis. The dimensions of the scanned domain were 30 × 120 mm at a spatial resolution of twelve points per wavelength (0.35 mm). At each spatial coordinate, a 60,000-point waveform was captured by a digital oscilloscope (DZ1054Z, Rigol Technologies Co., Ltd., Suzhou, China), and a Hilbert transform was applied to average the waveform. The spatial-peak instantaneous pressure in free water was derived from averaging the maximum of the Hilbert transform envelope over ten consecutive measurements at the beam maximum. Purified deionized water was used to fill the tank. The water temperature and dissolved oxygen levels were measured prior to the scan: 23.1 °C and 3.7 mg/L, respectively. The speed of sound in water was taken as 1491 m/s. The pulse timing parameters of the signal were identical to the pulse used in the study.

Independent sources of axial location error are the resolution uncertainty of the sampling step size ($\delta e_{\mathrm{sampling}}$), the intrinsic manufacturer reported motors error for the length of the travel distance ($\delta e_{\mathrm{motors}}$) and the error in visually determining the transducer’s exit plane ($\delta e_{\mathrm{visual}}$). The total error for the axial location is estimated by

$$\mathrm{\Delta }{e}_z=\sqrt{\underset{{\delta }_{\mathrm{sampling}}}{\underbrace{{\left(\mathit{dz}/2\right)}^2}}+\underset{{\delta }_{\mathrm{motors}}}{\underbrace{{\left(0.076\right)}^2}}+\underset{{\delta }_{\mathrm{visual}}}{\underbrace{{\left(\sigma /\sqrt{N}\right)}^2}}}=\pm 0.47\left[\mathrm{mm}\right]$$ 1

where $\mathit{dz}=0.35\left[\mathrm{mm}\right]$ is the measurement step size in the axial direction, $\sigma =0.953\left[\mathrm{mm}\right]$ is the standard deviation calculated from the difference in spatial peak location for N = 5 scans.

Lateral location errors are estimated by

$$\mathrm{\Delta }{e}_x=\sqrt{\underset{{\delta }_{\mathrm{sampling}}}{\underbrace{{\left(\mathit{dx}/2\right)}^2}}+\underset{{\delta }_{\mathrm{motors}}}{\underbrace{{\left(0.076\right)}^2}}}=\pm 0.19\left[\mathrm{mm}\right]$$ 2

where $\mathit{dx}=0.35\left[\mathrm{mm}\right]$ is the measurement step size in the $x$ and direction.

The axial position of the spatial peak is measured at a depth of $65\pm 0.5\left[\mathrm{mm}\right]$ from the exit plane of the transducer.

Estimations of the in situ acoustic parameters at the GPi were derived by sequentially derating free-water baseline measurements to account for skull transmission and brain tissue attenuation. Free-water measurements at 23 °C yielded a spatial peak pressure of 1.34 MPa, an I_sppa-w of 60.1 W/cm²_, and an I_spta-w of 3 W/cm². Transcranial pressure transmission (T_skull) was measured through two human skull specimens, targeting through the temporal squamosal suture to replicate the sonication conditions. To establish an upper bound exposure estimate, the maximum measured transmission coefficient from eight empirical experiments (T_skull = 0.41, range: 0.35–0.41). This corresponds to a 16.8% intensity transmission coefficient, consistent with published 350 kHz through-skull reported measurements at 350 kHz²⁸. Applying this coefficient yielded intermediate through-skull parameters: P_sp = 0.55 MPa, I_sppa-s = 10.1 W/cm², and I_spta-s = 0.51 W/cm². Subsequent derating using standard brain tissue attenuation models ($I_{\mathrm{GPi}}=I_w\cdot T_{\mathrm{skull}}^2\cdot {10}^{\alpha \cdot f\cdot d/10}\left[{\mathrm{W}/\mathrm{c}\mathrm{m}}^2\right]$) yielded estimated in situ values of I_sppa-GPi of 7.9 W/cm² and I_spta-GPi of 0.4 W/cm². Where a brain attenuation coefficient ($\alpha$) of $0.5{\mathrm{dB}\mathrm{c}\mathrm{m}}^{-1}{\mathrm{MH}\mathrm{z}}^{-1}$²⁹, a frequency ($f$) of $0.35\mathrm{MHz}$ and a dura to target depth (d) of 6 cm were employed. In situ estimated pressure was recovered from the derated intensity assuming an in-brain acoustic impedance of $Z_{\mathrm{brain}}=1.62\mathrm{MRayl}$³⁰, resulting in a peak in situ pressure amplitude of 0.5 MPa. The transcranial mechanical index $\left({\mathrm{M}\mathrm{I}}_{\mathrm{tc}}\right)$ was consequently estimated at 0.86. The cranial thermal index (TIC) was calculated according to IEC 62359:2010/AMD1:2017, where the time-average ultrasonic power of 0.54 W was derived by utilizing a Gaussian beam waist of 3.36 mm from the empirically measured 5.6 mm FWHM of pressure, and an equivalent aperture diameter $D_{\mathrm{eq}}$ of 8.5 cm. The resulting TIC value of 1.6 corresponded with a recommended maximum continuous sonication time of 80 min^29,31. We further estimated an extreme case where the effective aperture is 20% smaller due to an over-inflated water bladder. This potentially results in a higher energy density, which correlates with a TIC value of 2.0, maintaining the same permissible exposure limit.

Statistics and reproducibility

Means and standard deviations were used to summarize changes from baseline in AHRS–HCS and total DESA scores by treatment condition. The main analysis used a mixed-effects model including all post-baseline observations to estimate whether the mean change from baseline differed between TUS and sham. This analysis was intended to describe whether the estimated TUS–sham difference varied across follow-up time points and was not designed as a confirmatory test of delayed or sustained treatment effects. The model included treatment condition, visit, and intervention period as fixed effects and participant ID as a random intercept, assuming Gaussian error. Treatment effects are reported as estimated mean differences for TUS minus sham, with 95% confidence intervals and p-values. A two-sided p-value < 0.05 was considered statistically significant for the main analysis. Cohen’s d was calculated as the treatment-effect estimate from the mixed-effects model divided by the pooled standard deviation of change scores.

Exploratory time-specific between-group comparisons were conducted using a treatment-by-time mixed-effects model including the five pre-specified post-baseline time points: D, P, P1H, P1D, and P7D. This model included a subject-level random intercept to account for repeated measures and adjusted for the intervention period. Estimated mean differences were reported as TUS minus sham at each time point. To account for multiple comparisons across the five exploratory time-specific contrasts, nominal p-values were adjusted using the Holm stepdown procedure.

Analyses were conducted using RStudio 2023.06.1 and R 4.2.3. Individual de-identified data and statistical code are available from the corresponding author upon reasonable request.

Author contributions

W.Q. contributed to the conception and design of the study, secured funding, oversaw study execution, and drafted the manuscript. S.B. and A.G. performed the stimulation in patients; S.B. designed and fabricated the TUS system. A.G. validated the TUS system output and drafted the technical aspects of the manuscript. M.S. performed the statistical analyses. G.C. managed study administration and participant recruitment and conducted participant assessments. F.A. assisted with participant recruitment and data management. S.S. supervised the study’s technical aspects. D.G. supervised the study design, grant application, and overall execution. All authors critically reviewed and revised the manuscript and have read and approved the final version.

Data availability

The datasets generated and/or analyzed during the current study are not publicly available due to containing potentially sensitive or identifying clinical information about study participants, but are available from the corresponding author on reasonable request.

Code availability

The statistical analysis code used in this study is available online (published on Zenodo https://zenodo.org/records/21695356)³².

Declarations

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s44384-026-00076-3.