TMS vs. Focused Ultrasound

Two non-invasive ways to push on a circuit through an intact skull, one mature and shallow, one focal and deep. Companion page to TMS and FUS.

Teaching note·version 1.0·15 September 2026·not peer reviewed

Course notes, written to orient someone choosing between the two modalities. The citations and the numbers attributed to them have been checked against primary sources, but nothing here has been through external peer review, and the fast-moving parts, device clearances and the mechanism of ultrasonic neuromodulation above all, go stale quickly. Check anything you intend to rely on, and see what is verified and what is soft at the end for which parts of this page I would trust and which I would not.

Key takeaway: TMS induces an electric field that directly drives axons to threshold, so it produces action potentials you can measure in a muscle within milliseconds, but Maxwell's equations impose a depth versus focality trade that no coil design escapes. Low-intensity FUS delivers mechanical energy that a phased array can steer to a millimeter-scale spot anywhere in the brain, including targets TMS can never reach selectively, but it mostly biases excitability rather than forcing spikes, its mechanism is still contested, and the human evidence base is roughly where TMS was in the early 1990s.

Side by side

Property	TMS	Low-intensity FUS
Energy delivered	Electromagnetic. A 1 to 2.5 T pulse with a 50 to 100 microsecond rise time induces an electric field of roughly 100 V/m in cortex.	Mechanical. An acoustic pressure wave, typically 0.1 to 1.5 MPa peak rarefactional at the focus.
Carrier	dB/dt through a coil, no carrier frequency in the usual sense.	200 kHz to 1.5 MHz, with 250 to 700 kHz typical for transcranial work.
Skull	Effectively transparent. The skull does not perturb the magnetic field.	The hard problem. It attenuates, refracts, mode-converts and defocuses the beam, and it heats.
Coupling	None needed. Coil sits on the scalp, hair is fine.	Gel, a water bladder or a coupling cone. Hair degrades transmission and often has to be parted or wetted through.
Focal spot	Roughly two to three centimeters across for a figure-8 coil. In Deng's standardized metric the most focal figure-8 designs reach a half-value tangential spread of about 5 cm², and the field broadens with depth.	Roughly 2 to 5 mm lateral by 10 to 30 mm axial, a cigar aligned with the beam. Set by the wavelength and the aperture.
Depth	Gyral crowns and nearby cortex. Half-value depth is about 1.4 cm for a 70 mm figure-8, and at most 3.5 cm for any coil, since the induced field is provably zero at the center of the head.	Anywhere, including thalamus, amygdala, insula and brainstem, with the intensity peak at the focus rather than at the surface.
Effect on neurons	Suprathreshold. Depolarizes axons, preferentially at bends and terminals, and evokes spikes.	Mostly subthreshold. Biases excitability, apparently through mechanosensitive channels and membrane mechanics.
Direction of effect	Reasonably well mapped at the group level. Roughly, 1 Hz and cTBS suppress, 10 to 20 Hz and iTBS facilitate, though individual responses vary enough that a given subject may show the opposite.	Poorly mapped. Excitation versus suppression depends on duty cycle, pulse repetition frequency and target in ways that are not yet predictive.
Temporal precision	Sub-millisecond. Good enough for chronometry and virtual-lesion timing experiments.	Tens to hundreds of milliseconds per sonication. Fine for state changes, poor for single-event timing.
Direct readout	Motor evoked potentials give a same-session dose-response curve and a subject-specific threshold.	No equivalent. Effects are inferred from fMRI, EEG, evoked potentials or behavior.
Works inside an MRI	Awkward at best. Large artifacts, special hardware, rarely done.	Yes, routinely. Concurrent sonication and fMRI, and MR thermometry and MR acoustic radiation force imaging for targeting.
Targeting workflow	Neuronavigation to an anatomical or functional target, then set intensity relative to motor threshold.	Neuronavigation plus an acoustic simulation on a CT or MR-derived skull model to predict where the focus actually lands.
Main confound	A loud click and scalp and facial nerve stimulation. Sham coils are imperfect.	Skull-conducted sound at the pulse repetition frequency can drive auditory cortex and mimic a neural effect.
Regulatory status	Cleared for depression, OCD, smoking cessation and related indications. Thousands of clinical systems in service.	High-intensity thermal ablation is approved for tremor. Low-intensity neuromodulation remains investigational.
Cost and footprint	A clinical system is a capital purchase in the tens to low hundreds of thousands of dollars, and it rolls between rooms.	A benchtop single-element research rig is comparable. An MR-guided phased array helmet is a capital project.

Why the physics diverge at the skull

TMS ignores the skull and pays for it in depth Faraday's law, not conduction.
- Because the coil induces the field rather than injecting current, skull resistivity is irrelevant. That is exactly why TMS reaches threshold where tDCS cannot.
- The cost is that the induced field is set by the coil geometry alone, and it falls off steeply with distance. Making the hot spot deeper means making it wider, because you can only add depth by using larger current loops that are farther from the target.
- This is the depth versus focality trade, and it is a hard frontier rather than an engineering gap. Deng and colleagues simulated 50 commercial and experimental coils and found every one of them on the same curve, with no design beating the tradeoff set by figure-8 coils. It is also provable rather than merely empirical: at the quasi-static frequencies TMS uses, the induced field in a spherically symmetric conductor is always stronger at the surface than inside it, and exactly zero at the center.
FUS fights the skull and buys arbitrary depth Aberration correction, not geometry.
- Bone is faster, denser and far more absorbing than brain, and its thickness varies across the head. A wavefront crossing it arrives smeared unless each element of the array is delayed to compensate.
- With a CT-derived or MR-derived skull map and per-element phase correction, a large array can put a clean focus deep in the brain. Single-element transducers, which most neuromodulation labs use, cannot correct aberration and rely on simulation to estimate the resulting focal shift, which is commonly a few millimeters.
- Frequency is the central design knob and it cuts both ways. Lower frequency crosses the skull with less attenuation and less aberration, but the focal spot scales with wavelength, so you give up resolution to get transmission.
- Because the beam converges, intensity peaks at the focus rather than at the surface. That is the property TMS cannot match, and it is the whole reason FUS is interesting for subcortical targets.

A note on deep TMS

Deep TMS is the coil designer's answer to the depth problem, and it is worth being precise about what it does and does not buy, because the name invites a misreading.

How the coils work Slower decay, not a deep focus.
- The H-coil family, developed by Roth and Zangen and commercialized by Brainsway, wraps complex windings over a large area of the scalp and orients the current elements so their contributions summate at depth. The double cone coil does something similar with two large angled circular windings, and is the usual choice for leg motor cortex and medial structures such as anterior cingulate.
- What these designs change is the rate of decay, not the location of the maximum. You cannot create an interior focus with a quasi-static induced field the way an acoustic array creates one with a converging wave, so the field is still strongest at the surface and the deep target is stimulated only because everything above it is too.
- The trade is quantitative and steep. Against a 70 mm figure-8 coil, whose half-value depth is about 1.4 cm below the cortical surface, the modeled H coils gain 22 to 72 percent in depth while their tangential spread grows by 406 to 811 percent. Large double cone coils do somewhat better on both counts, reaching 2.5 to 3.1 cm with comparable or better focality, which is why they remain the pragmatic choice for medial targets.
What it is good for Clinically real, mechanistically blunt.
- Deep TMS carries its own clearances: the H1 coil for major depression in January 2013, the H7 for OCD in August 2018, which was the first clearance of any non-invasive device for OCD, and the H4 for smoking cessation in August 2020, the first for any addiction indication. This is established clinical practice, not a research curiosity.
- The costs are more scalp pain and facial muscle activation, and a larger induced field that plausibly raises seizure risk, consistent with reports of accidental seizure induction with H coils. The deeper problem is that the published rTMS safety limits were derived with roughly 70 mm figure-8 coils, so the large datasets that justify confidence there do not transfer cleanly to a coil with several times the field spread.
- For causal inference it is the wrong instrument. When the stimulated volume spans several centimeters of cortex plus whatever lies beneath it, an observed effect cannot be attributed to the deep structure you named in the methods.

This is the cleanest way to see what FUS is actually offering. Deep TMS buys somewhere between a third of a centimeter and a centimeter and a half of extra half-value depth while stimulating everything on the way in, against targets that can sit 6 cm or more from the scalp. FUS changes the shape of the problem, putting the intensity maximum at the target and leaving the overlying cortex comparatively alone. Depth and selectivity are separate axes, and deep TMS moves only one of them.

What actually happens to a neuron

This is the deepest difference between the two, and it is why the two literatures do not look alike.

TMS: a known transducer Field gradients drive axonal membranes.
- The induced electric field depolarizes axons where the field gradient along the fiber is largest, which in practice means bends, branch points and terminals rather than somata.
- Because the mechanism is understood, the dose has a natural unit. Intensity is expressed relative to each subject's motor threshold, and the motor evoked potential provides a direct check that the machine is doing something to the brain.
- Repetitive protocols then work through activity-dependent plasticity, with after-effects that resemble LTP and LTD and outlast the session.
FUS: several candidate transducers, none settled Mechanical coupling to the membrane.
- Leading candidates are direct gating of mechanosensitive channels such as Piezo1, TRP family channels and the two-pore potassium channels TREK-1, TREK-2 and TRAAK, along with acoustic radiation force, stable cavitation within the lipid bilayer, and changes in membrane capacitance.
- These are not mutually exclusive and their relative weight probably differs by cell type, which is one reason effects reverse sign across parameter sets that look similar on paper.
- Without a motor evoked potential equivalent, there is no quick positive control. A session can fail silently, and the field has not converged on a dose unit that transfers between labs.

The dials you actually set

TMS

Coil type: figure-8 for focality, double cone or H-coil for depth.
Intensity as a percent of resting or active motor threshold.
Pattern: single pulse, paired pulse for intracortical inhibition and facilitation, or repetitive trains.
Frequency and pattern for rTMS: 1 Hz, 10 to 20 Hz, or theta burst in continuous or intermittent form.
Pulses per session and sessions per day. Accelerated protocols compress a six-week course into days.
Coil orientation, which changes the induced current direction and materially changes the response.

FUS

Fundamental frequency, which sets skull transmission and focal size together.
Pressure or derated intensity at the focus, reported as spatial-peak pulse-average and spatial-peak temporal-average intensity.
Pulse repetition frequency and duty cycle, the two parameters most often blamed for flipping excitation to suppression.
Tone burst duration and total sonication duration, from a 100 ms burst during a task to tens of seconds for offline protocols.
Focal depth and steering, plus the aberration correction or simulation that justifies believing the focus is where you say.
Thermal budget for skull and brain, which constrains everything above.

Note the asymmetry. The TMS list is a short menu of protocols with published dose-response behavior. The FUS list is a continuous parameter space where the mapping to physiological effect is still being charted, which is why the ITRUSST consensus statements matter so much for reading the literature, the reporting standard from 2024 and the biophysical safety consensus from 2025.

Safety and who you cannot study

TMS Well characterized after four decades.
- Seizure is the serious risk and it is rare. A survey of active labs and clinics covering about 318,000 sessions from 2012 to 2016 found 24 seizures, roughly 0.08 per 1000 sessions overall, and fewer than one per 60,000 sessions when TMS was delivered within published guidelines to people without risk factors. Risk concentrates in people with epilepsy, brain lesions, certain medications or sleep deprivation, and rises when guidelines are exceeded.
- Common and minor: scalp pain, facial twitching, headache, and transient hearing effects from the click, which is why hearing protection is standard. Worth knowing, that same survey found high-frequency rTMS within guidelines no more likely to cause seizures than low-frequency or single and paired pulses, which contradicts a common assumption.
- Excluded: ferromagnetic hardware near the coil, cochlear implants, some stimulator hardware. Consensus safety and reporting guidelines are mature and widely followed.
FUS Good record so far, thinner denominator.
- The two limits to respect are thermal rise, which is often worst in the skull rather than the target, and mechanical effects from cavitation. The ITRUSST biophysical safety consensus puts concrete numbers on both for nonsignificant risk: a mechanical index at or below 1.9, and either a peak temperature rise no greater than 2 degrees C, a peak absolute temperature no greater than 39 degrees C, or a thermal dose within 2 CEM43 in brain, 16 CEM43 in bone and 21 CEM43 in skin. The authors are explicit that these are consensus levels, not established safety limits.
- Reported adverse events in human low-intensity studies are mild and mostly transient, headache, sleepiness, scalp sensation and neck pain, but total human exposure is orders of magnitude below TMS, so absence of rare events is not yet strong evidence.
- Excluded: skull plates, shunts and craniotomy hardware in the beam path, and beam paths that cross sinuses or thick, highly variable bone. Low skull density ratio is a known predictor of poor coupling for high-intensity work.

The confound each field has to argue about

Both modalities are noisy in a way that can masquerade as a neural effect, and in both cases the fix is the same in spirit, an active control that reproduces the peripheral sensation without the intended brain effect.

TMS delivers a loud, sharp click plus direct stimulation of scalp muscle and trigeminal afferents. Subjects can tell real from sham, and the click alone can entrain auditory responses in EEG.
FUS pulsed at an audible pulse repetition frequency drives the cochlea through bone conduction. Rodent work showed that a substantial part of what looked like direct neuromodulation was an indirect auditory response, which forced the field to adopt smoothed pulse envelopes, masking sound and deaf-animal controls.
Practical consequence for reading papers: ask what the control condition was before asking how large the effect was. This applies to both literatures, but it is more often unresolved in the FUS one.

Evidence and maturity

TMS A clinical tool with a research use.
- In clinical service since October 2008, when the first TMS device was cleared for major depression in adults who had not responded to at least one antidepressant in the current episode. Note that this is the actual indication, not treatment resistance in the strict sense. Later clearances cover OCD, smoking cessation, anxiety symptoms in adults with depression, and, in March 2024, adolescent depression as an adjunct for ages 15 to 21. The accelerated, imaging-targeted theta burst protocol from Stanford was cleared in September 2022.
- Large randomized trials, registries and meta-analyses exist. The open problems are effect size, responder prediction and the well-documented variability of after-effects between individuals, not whether the device does anything.
FUS A research tool with a clinical future.
- Human low-intensity work began in earnest in the 2010s with sensory and motor cortex studies, then extended to deep targets. Macaque work showing that 40 seconds of sonication changes brain activation for more than an hour afterwards, while controlling for auditory confounds, is a large part of why the field is optimistic.
- Studies are typically small, parameters vary widely, and replication across sites is the current bottleneck. Keep the ablation and blood-brain-barrier applications separate in your head when reading claims about approval status, since those are different intensity regimes with different regulatory histories.
- Worth noting, the high-intensity thermal use of the same physics is already approved and routine, for medication-refractory essential tremor since July 2016, for tremor-dominant Parkinson's since 2018, and later for the second side in essential tremor. The hardware, targeting and safety engineering have been through the clinic even though the low-intensity neuromodulation indication has not.

Choosing between them

Reach for TMS when

The target is cortical and reachable from the surface.
You need spikes, not a bias, for example a virtual lesion or a corticospinal probe.
You need millisecond timing relative to a stimulus or a response.
You want a per-subject dose readout, which means anything anchored to motor threshold.
The work is clinical, or has to be defensible to an IRB with minimal novel-device burden.

Reach for FUS when

The target is subcortical or deep, thalamus, amygdala, insula, striatum, brainstem.
You need the deep target modulated without modulating the cortex above it.
You want millimeter-scale spatial specificity, including distinguishing adjacent nuclei.
You want to stimulate and image at the same time inside an MRI.
You are pursuing something TMS cannot do at all, such as localized blood-brain barrier opening or sonogenetic approaches.

They are not really competitors. TMS is the right instrument for cortical causal questions and for a clinic today. FUS is the right instrument for deep causal questions and is the plausible non-invasive alternative to DBS for some indications tomorrow. The honest summary of the current state is that TMS knows what it does but cannot reach very far, and FUS can reach anywhere but is still establishing what it does.

Starting points in the literature

Every item below was checked against its PubMed record. Citations link to the publisher via DOI. The regulatory dates on this page were checked separately against FDA decision records and the corresponding manufacturer announcements.

Barker AT, Jalinous R, Freeston IL. Non-invasive magnetic stimulation of human motor cortex. Lancet, 1985, 1(8437):1106-7. The original demonstration. doi:10.1016/s0140-6736(85)92413-4
Roth Y, Zangen A, Hallett M. A coil design for transcranial magnetic stimulation of deep brain regions. Journal of Clinical Neurophysiology, 2002, 19(4):361-70. The origin of the H-coil. doi:10.1097/00004691-200208000-00008
Deng ZD, Lisanby SH, Peterchev AV. Electric field depth-focality tradeoff in transcranial magnetic stimulation, simulation comparison of 50 coil designs. Brain Stimulation, 2013, 6(1):1-13. The quantitative statement of the depth limit, and the source of the coil numbers used on this page. doi:10.1016/j.brs.2012.02.005
Huang YZ, Edwards MJ, Rounis E, Bhatia KP, Rothwell JC. Theta burst stimulation of the human motor cortex. Neuron, 2005, 45(2):201-6. doi:10.1016/j.neuron.2004.12.033
Lerner AJ, Wassermann EM, Tamir DI. Seizures from transcranial magnetic stimulation 2012-2016, results of a survey of active laboratories and clinics. Clinical Neurophysiology, 2019, 130(8):1409-16. The source of the seizure rates quoted above. doi:10.1016/j.clinph.2019.03.016
Rossi S, Antal A, Bestmann S, et al. Safety and recommendations for TMS use in healthy subjects and patient populations, with updates on training, ethical and regulatory issues. Clinical Neurophysiology, 2021, 132(1):269-306. The current consensus guideline. doi:10.1016/j.clinph.2020.10.003
Tyler WJ, Tufail Y, Finsterwald M, Tauchmann ML, Olson EJ, Majestic C. Remote excitation of neuronal circuits using low-intensity, low-frequency ultrasound. PLoS ONE, 2008, 3(10):e3511. An early demonstration of ultrasonic neuromodulation. doi:10.1371/journal.pone.0003511
Legon W, Sato TF, Opitz A, et al. Transcranial focused ultrasound modulates the activity of primary somatosensory cortex in humans. Nature Neuroscience, 2014, 17(2):322-9. Beam resolution of 4.9 mm lateral by 18 mm axial, for scale. doi:10.1038/nn.3620
Guo H, Hamilton M, Offutt SJ, et al. Ultrasound produces extensive brain activation via a cochlear pathway. Neuron, 2018, 98(5):1020-30. doi:10.1016/j.neuron.2018.04.036
Sato T, Shapiro MG, Tsao DY. Ultrasonic neuromodulation causes widespread cortical activation via an indirect auditory mechanism. Neuron, 2018, 98(5):1031-41. With the paper above, the auditory confound problem. doi:10.1016/j.neuron.2018.05.009
Verhagen L, Gallea C, Folloni D, et al. Offline impact of transcranial focused ultrasound on cortical activation in primates. eLife, 2019, 8:e40541. The 40 second sonication with effects lasting over an hour. doi:10.7554/eLife.40541
Folloni D, Verhagen L, Mars RB, et al. Manipulation of subcortical and deep cortical activity in the primate brain using transcranial focused ultrasound stimulation. Neuron, 2019, 101(6):1109-16. Amygdala and anterior cingulate in macaques. doi:10.1016/j.neuron.2019.01.019
Blackmore J, Shrivastava S, Sallet J, Butler CR, Cleveland RO. Ultrasound neuromodulation, a review of results, mechanisms and safety. Ultrasound in Medicine and Biology, 2019, 45(7):1509-36. doi:10.1016/j.ultrasmedbio.2018.12.015
Martin E, Aubry JF, Schafer M, Verhagen L, Treeby B, Butts Pauly K. ITRUSST consensus on standardised reporting for transcranial ultrasound stimulation. Brain Stimulation, 2024, 17(3):607-15. doi:10.1016/j.brs.2024.04.013
Aubry JF, Attali D, Schafer ME, et al. ITRUSST consensus on biophysical safety for transcranial ultrasound stimulation. Brain Stimulation, 2025, 18(6):1896-905. Read this and the one above before designing or reviewing a FUS protocol. doi:10.1016/j.brs.2025.10.007

What is verified and what is soft

Written on 15 September 2026, and true of the page as of that date.

Checked against primary sources

Every reference in the list above, confirmed against its PubMed record for authors, journal, year, volume, pages and DOI. All fifteen are real and are what this page says they are.
All of the coil geometry numbers, taken from the full text of Deng, Lisanby and Peterchev: the 1.4 cm half-value depth of a 70 mm figure-8, the 22 to 72 percent depth gain and 406 to 811 percent spread increase for H coils, the 2.5 to 3.1 cm reached by large double cone coils, the 0.9 to 3.5 cm range across all 50 designs, the 5 cm² best-case tangential spread, and the ceiling at half the brain radius.
The proof sketch behind that ceiling, that at quasi-static frequencies the induced field in a spherically symmetric conductor is strongest at the surface and zero at the center.
The seizure statistics, from Lerner, Wassermann and Tamir: 24 seizures in about 318,000 sessions, roughly 0.08 per 1000, fewer than one per 60,000 within guidelines and without risk factors, and high-frequency rTMS no riskier than low-frequency under those conditions.
The ITRUSST safety levels quoted verbatim from the consensus: mechanical index 1.9, a 2 degree C rise or 39 degree C absolute, and thermal doses of 2, 16 and 21 CEM43 in brain, bone and skin.
The specific experimental results attributed to particular papers: Legon's 4.9 mm by 18 mm beam, Verhagen's 40 second sonication with effects beyond an hour, and Folloni's amygdala and anterior cingulate targeting in macaques.
Every regulatory date, against FDA decision records and the matching manufacturer announcements. One caveat on method: the FDA's own database was not directly reachable from the machine this was written on, so the decision documents were read as surfaced alongside the manufacturer and trade-press announcements, which agreed in every case.

Soft, and worth checking before you rely on it

Most of the engineering ranges in the side-by-side table, the 1 to 2.5 T pulse, the 50 to 100 microsecond rise time, the 100 V/m field, the 0.1 to 1.5 MPa pressures, the frequency and focal-spot ranges beyond Legon's measured figures, and the duty cycle and pulse repetition conventions. These are conventional textbook values, given as typical rather than as measurements, and no single source on this page establishes them.
The cost and footprint row. That is market impression, not a sourced figure, and it is the least reliable line on the page.
Everything about the mechanism of ultrasonic neuromodulation. The candidate transducers named are all genuinely in the literature, but their relative importance is unsettled, and any impression this page gives about which matters most is my judgment rather than a finding.
The claim that duty cycle and pulse repetition frequency are what flip excitation to suppression. Widely believed, repeatedly reported, not established.
The single-element focal shift of a few millimeters, which varies with skull, target and frequency far more than a single number suggests.
The direction-of-effect summary for TMS. It holds at the group level, and the page says so, but it is a simplification of a literature with well-documented individual variability.
The adverse event list for low-intensity ultrasound, which reflects the review literature in general terms rather than a specific tabulation.
That low-intensity ultrasonic neuromodulation remains investigational. True as far as I know on the date above, but device clearances move, and this is the sentence most likely to be wrong first.
The date of the anxiety-symptoms indication, deliberately omitted because the indication is confirmed and the date was not.
The framing: that the two are complements rather than competitors, that the FUS evidence base sits roughly where TMS did in the early 1990s, and the whole of the decision guide. Those are arguments, not facts, and a reasonable person in the field could disagree with any of them.

Corrections are welcome, and the parts above are listed in roughly the order I would expect them to need fixing.