Transcranial Focused Ultrasound (tFUS)

• tFUS is the transcranial delivery of focused acoustic waves through the intact skull for neuromodulation and therapy, combining low-intensity stimulation with high-precision targeting.
• It spans diverse regimes such as low-intensity nonthermal neuromodulation, MRI-guided thermal ablation, and microbubble-mediated blood-brain barrier opening, each with distinct acoustic workflows.
• Skull-aware modeling and advanced transducer architectures are critical for accurate phase correction, safety assessment, and achieving millimeter to micron-scale spatial resolution.

Transcranial Focused Ultrasound (tFUS) is the transcranial delivery of focused acoustic pressure waves through the intact skull to modulate neural activity or deliver therapy noninvasively. In contemporary usage, the term spans low-intensity neuromodulation protocols and related transcranial focused-ultrasound interventions such as MRI-guided thermal ablation and microbubble-mediated blood-brain barrier opening. Across these regimes, the defining advantages are millimeter-scale spatial resolution, access to cortical and subcortical targets, and the possibility of patient-specific wavefront control; the defining technical constraint is the skull, whose thickness, porosity, curvature, attenuation, and elastic behavior distort and attenuate the beam and make case-specific modeling central to targeting accuracy and safety (Ai et al., 2016, McDannold et al., 2019, Angla et al., 2022).

1. Modality boundaries and operating regimes

Low-intensity tFUS is described as focused ultrasound applied through the skull to modulate neuronal excitability without surgery, typically with non-thermal mechanical bioeffects rather than tissue destruction. This regime is explicitly contrasted with high-intensity focused ultrasound used for thermal ablation, while TcMRgFUS denotes MRI-guided therapeutic systems that focus sound through the skull to a small region under MRI guidance and are clinically approved to thermally ablate regions of the thalamus (Tyler, 1 Apr 2026, Liu et al., 2022).

Within this broader family, different therapeutic objectives correspond to distinct acoustic workflows. TcMRgFUS thermal ablation has been implemented clinically on the ExAblate Neuro platform with a 1024-element hemispherical phased-array transducer operating at 660 kHz inside a 3T MRI scanner, with CT-based phase correction, water coupling, and active cooling to reduce heat at the scalp and skull surface (McDannold et al., 2019). By contrast, microbubble-assisted blood-brain barrier opening uses lower-pressure pulsed insonation and passive detection of microbubble acoustic emissions; a recent all-CMUT platform combined five transmitting elements with one receiving element at 700 kHz to perform both therapy and broadband sensing in rats (Kilinc et al., 24 Apr 2026).

The category also includes high-precision acoustic descendants that retain the acoustic modality while departing from standard transcranial piezoelectric focusing. Optically-generated focused ultrasound (OFUS) produced a 15 MHz transcranial focus in mice with measured lateral resolution of 83 $\mu\text{m}$ through skull, while the tapered fiber optoacoustic emitter (TFOE) was presented as a non-genetic, fiber-delivered optoacoustic source with about 20 $\mu\text{m}$ confinement for single-neuron and subcellular stimulation ex vivo (Li et al., 2022, Shi et al., 2020). These systems are not direct replacements for conventional human tFUS, but they define an important extension of the field toward ultrahigh spatial specificity.

2. Physical basis and biophysical mechanisms

The physical action of low-intensity tFUS is framed primarily in mechanical terms. Reported mechanisms include acoustic radiation force, shear stress, lipid-bilayer deformation, and activation of mechanosensitive ion channels including Piezo1, TRAAK, TREK, and TRP channels. The literature summarized for ultrasonic brain-computer interfaces also cites the Bilayer Sonophore model and the Neuronal Intramembrane Cavitation Excitation model, both of which treat ultrasound pressure cycles as perturbations capable of changing membrane capacitance and generating displacement currents (Tyler, 1 Apr 2026).

Temporal structure is treated as mechanistically important rather than incidental. Pulse repetition frequency, duty cycle, and pulse duration are repeatedly identified as key control parameters, and the review of uBCIs reports a general tendency in which high duty cycle or high PRF is more often excitatory, whereas low duty cycle or low PRF is more often inhibitory; the examples given are $>50\%$ duty cycle or $>500$ Hz PRF for excitatory protocols, and $<30\%$ duty cycle or $<100$ Hz PRF for suppressive protocols (Tyler, 1 Apr 2026). This does not constitute a universal law, but it establishes that outcome polarity is parameter dependent.

The skull is the dominant perturbing medium. It introduces reflection, refraction, attenuation, phase aberration, and, in cortical or shallow-target geometries, significant mode conversion into shear waves. For shallow cortical propagation, the standard fluid-skull approximation was shown to be inadequate: ignoring shear waves produced an average $\sim 40\%$ overestimation of intracranial acoustic pressure and a maximum mean deviation of focal area of 125% at oblique incidence, whereas the solid model was more stable under small skull-transducer misregistrations (Gao et al., 2023). Long-pulsed measurements across human skull specimens further showed that thickness variation within a local region can be more consequential than mean thickness alone, with some regions exhibiting severe focal shift, focal-area expansion, and loss of a meaningful focus at 650 kHz or 1000 kHz (Li et al., 2024).

A separate mechanistic uncertainty concerns what biological endpoint is being perturbed. In human tFUS-fMRI, one proposed interpretation is direct neuronal modulation; another is mechanical influence on local microvasculature that then generates the BOLD signal. The human 3T/7T study did not rule out the vascular account, although it noted prior behavioral and electrophysiological findings consistent with a neuronal mechanism (Ai et al., 2016). This leaves the field with a genuine mechanistic plurality rather than a settled single-pathway account.

3. Skull-aware modeling and treatment planning

Case-specific simulation is a foundational requirement in tFUS because the skull is highly attenuating, strongly aberrating, and variable across individuals and across anatomical regions within one skull. The major planning tasks repeatedly identified are phase and amplitude correction, thermal prediction, and safety assessment for cavitation, standing waves, and off-target exposure (Angla et al., 2022).

The quantitative uncertainty in current skull models remains substantial. In a cross-comparison of five k-Wave skull parameterization strategies across 19 regions of interest from seven human skull specimens and three frequencies, mean peak-pressure errors ranged from 20% to 31%, intensity errors from 41% to 77%, $-6$ dB focal-volume errors ranged from 11% to 67%, and focal-position discrepancies were typically several millimetres. Simulations also generally predicted smaller insertion losses than measured, implying systematic underestimation of skull-related attenuation and overestimation of transmitted intracranial exposure (Li et al., 8 Jun 2026). This establishes that present planning pipelines can reproduce gross beam patterns without yet delivering uniformly reliable quantitative dosimetry.

Open-source and accelerated planning systems have emerged to address this translational bottleneck. BabelBrain integrates MRI, optional CT or zero-echo time MRI, transcranial acoustic modeling with BabelViscoFDTD, and bioheat simulation, and reports total runtime around 10 minutes or less on capable hardware. In its internal comparison of Hounsfield-unit mapping laws, the Webb-Marsac method with $q_c = 3$ produced the smallest average distance to literature transmission values, at 0.021 (Pichardo, 2023). A plausible implication is that practical neuromodulation workflows will increasingly rely on calibrated, software-mediated forward modeling rather than heuristic placement alone.

A parallel development is CT substitution and learned surrogate simulation. MRI-to-synthetic-CT methods based on 3D patch-based cGANs were shown to preserve clinically relevant skull metrics for planning: in one study, synthetic CT and real CT had Pearson correlations of 0.94 for skull density ratio, 0.92 for skull thickness, and 0.98 for number of active elements, while the distance between peak focal locations remained below 1.3 mm across simulation cases (Liu et al., 2022). A later fully CT-free framework combining MRI-derived synthetic CT with k-Wave and accelerated angular-spectrum pipelines reported sub-millimeter transverse targeting deviation, focal shape consistency with FWHM about 3.3–3.8 mm, normalized pressure error below 0.2, and runtime reduction from about 3320 s to 187 s or 34 s depending on the solver (Gao et al., 11 Jul 2025). At larger scale, TFUScapes and DeepTFUS introduced 2,500 high-resolution 3D skull simulations from 125 subjects, using a transducer-aware U-Net surrogate trained on k-Wave outputs (Srivastav et al., 19 May 2025).

4. Transducer architectures and wavefront control

tFUS hardware spans single-element bowls, phased arrays, CMUT arrays, and acoustic holographic interfaces. In clinical TcMRgFUS, the canonical architecture remains a hemispherical phased array with CT-based aberration correction, exemplified by the 1024-element ExAblate Neuro system at 660 kHz (McDannold et al., 2019). For non-human primates, a purpose-built MR-guided randomized sparse spherical-cap array with 128 elements at 650 kHz, radius of curvature 72 mm, and opening diameter 103 mm was optimized for macaque S1 targeting; hydrophone measurements estimated usable steering ranges of about $\pm 20$ mm axially and $\mu\text{m}$0 mm transversely, with the strongest free-field grating lobe around $\mu\text{m}$1 dB relative to the focus (Chaplin et al., 2017).

Broadband therapy-and-sensing integration is represented by CMUT platforms. The half-ring all-CMUT array for BBB opening used five transmitters and one receiver, a 5-cm geometric focus, and phase inversion to suppress device-generated nonlinear harmonics. In water, phase inversion reduced the fundamental by 58 dB and the third harmonic by 36 dB; across system conditions, it produced a 7–20 dB enhancement in effective dynamic range, improving separation of microbubble emissions from transmit-side nonlinear contamination (Kilinc et al., 24 Apr 2026). The significance is not merely hardware substitution: it is the move toward tFUS systems that sense bubble dynamics and potentially regulate themselves in real time.

Wavefront engineering is also moving away from purely electronic beamforming. Gradient-descent volumetric holography using a modified mixed-domain method at 444 kHz replaced time-reversal-based lens design with an explicitly optimized 3D target-field objective, achieving average convergence in seven iterations and total design time of about 44 minutes for targets including the insula, hippocampus, caudate, and amygdala (Sallam et al., 2024). Subject-specific volumetric holography then extended this logic to skull-conforming acoustic holographic lenses that jointly encode individualized phase correction and a conformal coupling layer; across 20 cases, mean targeted sonicated volume fraction was reported as $\mu\text{m}$2, with experimental validation in an ex vivo human skull at 675 kHz (Cengiz et al., 23 Apr 2026). At higher frequencies, frequency-domain topology optimization with HASA and ADAM was proposed because simple phase-to-thickness conversion breaks down in acoustically thick holograms; reported improvements included SSIM changes from 0.25 to 0.77 at 0.5 MHz and PSNR gains from 14.83 dB to 18.88 dB at 1 MHz (Dash, 4 Jun 2026).

5. Experimental scope: from single neurons to human circuits

Human neuromodulation studies have used tFUS both as a perturbational tool and as a probe of distributed circuits. In the first human tFUS-BOLD demonstrations, 3T stimulation of the primary motor cortex hand knob produced no statistically significant group-level activation, yet 3 of 6 participants showed a very focal BOLD response in the sensorimotor region aligned with the ultrasound focus. A 7T pilot targeting the left head of caudate in one volunteer showed a focal BOLD response in the targeted subcortical region (Ai et al., 2016). A later methodological roadmap argues that this combination of noninvasiveness, depth access, and millimeter-scale focality makes tFUS especially attractive for consciousness research, where causal perturbation of small cortical patches or deep nuclei such as the pulvinar, amygdala, or basal forebrain is difficult with TMS, tDCS, or tACS (Freeman et al., 11 Jul 2025).

Microbubble-mediated BBB opening is the most developed non-neuromodulatory low-intensity application in the supplied literature. In rats, the CMUT platform achieved spatially localized BBB opening confirmed by T1-weighted MRI, and DCE-MRI permeability mapping showed pressure-dependent scaling of $\mu\text{m}$3: $\mu\text{m}$4 for the high-pressure hemisphere, $\mu\text{m}$5 for the low-pressure hemisphere, and $\mu\text{m}$6 in the unsonicated control (Kilinc et al., 24 Apr 2026). Time-resolved acoustic spectra also tracked microbubble arrival, circulation, and washout, which is directly relevant to feedback-controlled delivery.

Thermal ablation remains the most mature therapeutic endpoint. In 40 clinical TcMRgFUS procedures with delayed MRI follow-up, skull lesions were observed after 16 treatments, corresponding to 40%, and lesion onset was associated with maximum acoustic energy thresholds of 18.1–21.1 kJ and total acoustic energy thresholds of 97–112 kJ in the abstract, with 97–122 kJ reported in the results section. Acoustic energy predicted whether lesions would occur, whereas acoustic-plus-thermal simulations better predicted lesion location and extent, with excellent agreement to delayed thin-slice T2-weighted MRI (McDannold et al., 2019). This makes the skull not merely a transmission barrier but a therapeutic dose-limiting structure.

At the opposite spatial extreme, optoacoustic systems have reduced acoustic stimulation volumes from millimeters to tens of microns. TFOE generated near-field pressure of 56.7 kPa, enabled whole-cell patch clamp during acoustic stimulation, and showed selective activation of neurons separated by $\mu\text{m}$7, with only the intended cell responding (Shi et al., 2020). OFUS used a soft optoacoustic pad to generate a single-cycle focused stimulus through mouse skull, with measured lateral resolution of 83 $\mu\text{m}$8, c-Fos-positive fraction of $\mu\text{m}$9 versus $>50\%$0 in controls, and contralateral hind-limb EMG responses after motor-cortex targeting (Li et al., 2022). These systems suggest that the mechanistic study of ultrasound neuromodulation can now be performed at cellular and subcellular scales that conventional transcranial beams do not reach.

6. Safety, limitations, and unresolved questions

Safety assessment in tFUS involves both exposure metrics and delayed biological endpoints. A standard metric is the mechanical index,

$>50\%$1

and the uBCI review summarizes guidance including $>50\%$2, $>50\%$3, and intracranial temperature rise $>50\%$4. It also notes that low-intensity human studies often use roughly 0.5 to 35 W/cm$>50\%$5, while skull attenuation can reduce intensity by up to 90%, making free-field and in situ estimates non-equivalent (Tyler, 1 Apr 2026). In OFUS, the reported mechanical index was 0.5 and estimated transcranial temperature rise during the longest stimulation was $>50\%$6 K (Li et al., 2022).

Clinical safety is not exhausted by acute monitoring. In TcMRgFUS ablation, no skull damage was visible on 24-hour post-treatment MRI, yet delayed T2-weighted follow-up at 3–15 months identified asymptomatic skull lesions in 16 of 40 treatments, with thin-slice imaging providing the best detection (McDannold et al., 2019). Concurrent MRI introduces additional engineering constraints: in a 7T test, powered 0.5 MHz transducers warmed by about $>50\%$7 or $>50\%$8 over 20 minutes under the reported conditions, and susceptibility artifact from PZT could be reduced by moving the transducer about 4 cm from a phantom or by careful shimming, though not eliminated (Ai et al., 2016).

Several technical controversies remain open. One is acoustic modeling fidelity: current skull models can capture gross intracranial beam patterns yet still misestimate attenuation, focal volume, and location by margins large enough to matter experimentally and clinically (Li et al., 8 Jun 2026). Another is whether shallow cortical tFUS can be planned with fluid-skull assumptions; the evidence argues that it cannot when oblique incidence and shear-wave effects dominate (Gao et al., 2023). A third is interpretational: neuromodulatory effects may reflect direct neuronal perturbation, vascular mediation, or mixed pathways, and human studies also need to address audible hardware confounds through masking sounds, modulation envelopes, or unfocused controls (Ai et al., 2016, Freeman et al., 11 Jul 2025).

The current trajectory of the field is toward more anatomically adaptive, feedback-capable, and radiation-free systems. MRI-only synthetic-CT planning, all-CMUT therapy-and-sensing arrays, subject-specific volumetric holography, and bidirectional ultrasonic interfaces are all explicitly proposed as next steps in the cited literature (Gao et al., 11 Jul 2025, Kilinc et al., 24 Apr 2026, Cengiz et al., 23 Apr 2026, Tyler, 1 Apr 2026). This suggests that the future of tFUS is likely to depend less on any single transducer class than on the integration of skull-aware modeling, adaptive wavefront control, and quantitative monitoring into a unified transcranial acoustics platform.