Pith. sign in

REVIEW 3 major objections 4 minor 23 references

tFUSOperator: a coordinate-aware neural operator that maps free-field pressure, skull anatomy, and transducer parameters to the intracranial acoustic field, claiming the first operator-based formulation of transcranial focused ultrasound pr

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-04 19:50 UTC pith:V4UTYZNS

load-bearing objection Solid domain contribution: first operator-learning surrogate for tFUS with held-out skull generalization, but the MR pathway is underspecified and the empirical base is thin (13 simulated skulls, 3 test). the 3 major comments →

arxiv 2608.01839 v1 pith:V4UTYZNS submitted 2026-08-03 cs.LG cs.NAmath.NA

tFUSOperator: Operator Learning for Transcranial Focused Ultrasound Digital Twins

Minjee Seo , Haris Ghafoor , Minju Seol , Seonaeng Cho , Kyungho Yoon This is my paper
classification cs.LG cs.NAmath.NA MSC 68T0792C55
keywords transcranial focused ultrasoundneural operatordigital twincoordinate-aware attentionMR versus CTk-Waveintracranial pressure fielddeep learning surrogate
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper claims that transcranial focused ultrasound field prediction can be cast as an operator learning problem, replacing fixed-grid voxel regression with a coordinate-aware neural operator that maps free-field pressure, skull anatomy, and treatment parameters to the intracranial pressure field within a shared physical coordinate frame. The authors introduce tFUSOperator, a Transformer-based latent operator, and report that it generalizes across skulls: about 90% Dice on seen skulls with held-out transducer positions and about 72% Dice on entirely unseen skulls, with peak localization within roughly 3 mm and peak-magnitude error under 2%. They also show that magnetic resonance (MR) input performs nearly as well as computed tomography (CT), which matters because MR avoids ionizing radiation. If these results hold, tFUSOperator offers a radiation-free, interactive path to patient-specific digital twins, running 5.6e4 times faster than the k-Wave numerical solver.

Core claim

The central claim is that the skull-mediated acoustic operator G mapping free-field pressure, skull anatomy, and treatment parameters to the intracranial peak-pressure field can be approximated by a neural operator without fixing the output to a grid. tFUSOperator uses coordinate-decoupled attention: keys are generated purely from physical coordinates, while values combine coordinate embeddings with modality features, so attention is driven by where points are in space rather than by learned correlations. The decoder queries the latent representation at focal coordinates, making the model resolution-flexible. On unseen transducer positions on known skulls, the model reaches about 90% Dice an

What carries the argument

Coordinate-decoupled cross-attention in a latent Transformer neural operator. Free-field and skull volumes are tokenized and paired with normalized millimeter coordinates via sinusoidal positional encoding; keys come from coordinates only, while values concatenate coordinate and feature embeddings. This separates where the model attends (physical location) from what it reads (modality content), allowing a single model to integrate heterogeneous inputs across different physical domains and to reconstruct the field at arbitrary focal coordinates through the decoder query.

Load-bearing premise

The k-Wave simulations built from HU-segmented CT of 13 subjects at 1 mm isotropic resolution are treated as ground truth, and MR is assumed to encode enough bone structure to serve as a substitute for CT.

What would settle it

Run the model on a new cohort of at least 10 skulls with ground truth from higher-resolution k-Wave (0.5 mm) or from hydrophone measurements on ex-vivo skulls; if the unseen-skull Dice falls below roughly 50% or peak distance exceeds 5 mm, the claimed cross-skull generalization would be contradicted.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • A single tFUSOperator model handles multiple transducer placements and frequencies (250, 400, 500 kHz) by conditioning on position and orientation, so changing treatment parameters does not require retraining.
  • On seen skulls with held-out transducer positions, the model achieves ~90% Dice and sub-1.4 mm peak localization, indicating reliable interpolation over transducer setups.
  • On entirely unseen skulls, peak-magnitude error stays within ~2% even as Dice drops to ~72%, suggesting the operator transfers physical intensity information across anatomies.
  • MR input yields essentially the same Dice, peak distance, and peak difference as CT, supporting radiation-free imaging in a digital-twin pipeline.
  • At 2.16 ms per prediction versus 122 s for k-Wave, the model enables repeated, interactive field estimation needed for a tFUS digital twin.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: the coordinate-query decoder should allow resolution-flexible inference, so the same trained weights could reconstruct fields at finer than 1 mm spacing; testing this on unseen skulls would show whether focal-shape Dice improves without retraining.
  • Beyond the paper: coordinate-only keys suggest the model might transfer across imaging modalities beyond CT and MR (e.g., synthetic anatomical skull maps), since attention is anchored to physical millimeters rather than to tissue labels.
  • Beyond the paper: because the ground truth is simulated at 1 mm isotropic resolution, true peak pressure values could be underestimated; a validation against ex-vivo hydrophone measurements would test whether the reported ~2% magnitude error reflects simulator bias or genuine model accuracy.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes tFUSOperator, a coordinate-aware neural operator for transcranial focused ultrasound simulation. Instead of voxel-to-voxel regression, it learns an operator mapping the free-field pressure, skull anatomy (CT or MR), and transducer parameters to the intracranial peak-pressure field, using a shared physical coordinate frame. The model is a latent Transformer with dual-modality encoder, DiT-style conditioning on transducer position and direction, and coordinate-query decoder. Training data are k-Wave simulations generated for 13 CT-derived skulls at three frequencies and 300 transducer placements per skull, with skull-level splitting into 8/2/3 train/validation/test. On seen skulls with unseen transducer positions the paper reports about 90% Dice, 1.3 mm peak localization error, and about 1–2% peak magnitude difference; on unseen skulls it reports about 72% Dice and about 3 mm peak localization error. MR input is claimed to perform on par with CT, and inference is claimed to be 5.6e4 times faster than k-Wave.

Significance. If the central claims hold, the paper makes a useful contribution: it is the first operator-based formulation of tFUS field prediction, it provides a physically structured coordinate-aware attention mechanism rather than a fixed-grid regression, and it explicitly evaluates generalization to held-out skulls. The code release and skull-level split are strengths. However, two load-bearing pieces are not yet established: the MR-input pathway is not described at all, and the unseen-skull generalization claim rests on only three test subjects without per-subject statistics. The absence of any comparison with previously published deep-learning tFUS surrogates also makes it difficult to assess the claimed advantage over the prior art cited in the introduction.

major comments (3)
  1. [§2.2, §3.1, Table 1] The MR input pathway is entirely unspecified. Section 2.2 defines h_sk as a 'CT- or MR-derived function' and describes only CT/HU segmentation for the simulator. Nowhere in §3.1 or elsewhere is it stated how MR volumes are acquired, co-registered to the CT/simulation space, intensity-normalized, or converted to acoustic properties. Since MR does not directly provide bone density, a meaningful MR-based h_sk requires a defined surrogate mapping (e.g., synthetic-CT synthesis, registered MR with density scaling, or a learned modality mapping). Without this, the MR rows of Table 1 and Figure 2 cannot be reproduced, and the claim that MR is 'on par with CT' is unverifiable. If the MR inputs are, in fact, derived from CT in some way, that must be stated explicitly; otherwise the 'radiation-free' claim and the MR-parity conclusion rest on an unstated assumption.
  2. [§3.1, Table 1, §4] The generalization-to-unseen-skulls claim is based on only 3 test skulls. Table 1 reports means and standard deviations without specifying whether the standard deviation is over the 300 transducer positions per skull, over the 3 skulls, or over a pooled set. This matters because a single atypical test skull can dominate the average. The paper should report per-skull Dice, peak distance, and peak difference, as well as the range across the three test skulls. With only three subjects, the claim of cross-skull transfer needs more careful qualification than 'generalizes to new skull geometry.' At minimum, give individual skull results and confidence intervals.
  3. [§3.1, §3.2] No comparison is made to any existing deep-learning tFUS surrogate, despite the introduction citing several (Refs. [7], [13], [15], [16], [17], [18]). The contribution is framed partly by contrast with 'voxel-to-voxel regression' methods, and the speedup of 5.6e4 over k-Wave is useful, but the reader cannot tell whether the proposed operator model outperforms, matches, or underperforms earlier DL surrogate models on the same or similar data. A fair baseline comparison—at least on seen-skull/unseen-position accuracy and inference time—would substantially strengthen the performance claims.
minor comments (4)
  1. [§2.1, Eq. (1)] The notation F(Ω) is used for 'the space of scalar functions,' but the functional setting (e.g., L2, Sobolev regularity) is not specified. This is acceptable for a practitioner-oriented paper, but a brief clarification would help operator-learning readers.
  2. [§3.2] The speedup number is computed as 122 s / 2.16 ms ≈ 5.65e4, which is correct. However, the statement 'a 5.6×10^4-fold speedup' would be more transparent if the hardware and batch size used for the k-Wave comparison were given (e.g., same GPU, single sample).
  3. [§3.1, Evaluation Metrics] The Dice metric is said to measure overlap of FWHM regions, but the thresholding procedure (global vs. per-field, interpolation, discretization) is not fully specified. Please state the exact definition used.
  4. [§4] The Discussion mentions that the coordinate-query decoder supports arbitrary spatial resolution but says this has not been exploited. This is an honest limitation, but the phrasing 'naturally supports prediction at arbitrary spatial resolutions' should not be over-read; no experiment demonstrates resolution-generalization.

Circularity Check

0 steps flagged

No significant circularity; held-out evaluation and independent k-Wave ground truth keep the operator-learning claim self-contained. The only self-citation is for data generation, not for the predicted result.

full rationale

The central claim is that a trained neural operator G_theta maps free-field pressure, skull anatomy, and treatment parameters to intracranial peak pressure (Eq. 1). Training uses k-Wave simulated pairs from 13 skulls; evaluation is on held-out transducer positions and three entirely unseen skulls, so Dice, peak distance, and peak difference are out-of-sample, not fitted constants. The only self-citation is [16], used for the data-generation pipeline: 'We follow the simulation framework of [16] to generate paired free-field and intracranial fields and to incorporate skull images as an additional input modality.' This is prior work by the same authors, but it supplies a simulation benchmark (k-Wave) rather than the operator result itself; the paper's speed and generalization claims are evaluated against that benchmark and are not forced by construction. The MR input pathway is underspecified — no MR acquisition, registration, intensity normalization, or CT-to-MR mapping is given, so the reported MR parity is difficult to reproduce and could in principle depend on an unstated CT-derived proxy. However, the paper does not state that MR volumes are CT-derived or define MR predictions in terms of CT, so no specific circular reduction can be exhibited from the text. Section 4 openly notes the 13-subject limitation, which is a data-scale concern, not circularity. Overall, no equation or fitted parameter is equivalent to the reported prediction; score 2 reflects the minor self-citation, not partial circularity.

Axiom & Free-Parameter Ledger

4 free parameters · 3 axioms · 0 invented entities

The central claim rests on three classes of unverified inputs: simulator ground truth, MR-anatomy mapping, and representativeness of the 13-skull sample. All quantitative results are produced by a fitted neural network; the fitting itself is standard and validated out-of-sample, so it is not a hidden circularity. The paper introduces no new physical entities.

free parameters (4)
  • Architecture hyperparameters (d=384, M=512 latent tokens, L=8 blocks, 6 heads, 32-channel stem) = hand-chosen; no sensitivity study
    Capacity and depth are chosen without derivation and affect all reported metrics.
  • Loss balancing log-variances for MSE and soft Dice = learned during training
    The two loss terms are combined by homoscedastic uncertainty weighting, so the effective loss weights are fitted rather than fixed.
  • All trainable weights of the encoder, processor, and decoder = learned on 8 training skulls
    The model is a fitted surrogate; the Table 1 numbers are produced by these weights. This is standard supervised fitting, not a hidden circular constant.
  • Simulation dataset design (3 frequencies, 300 transducer placements per skull, 56^3 ROI, 1 mm spacing) = hand-selected
    These protocol choices determine what the surrogate learns and what the reported Dice and speedups mean; no sensitivity analysis is provided.
axioms (3)
  • domain assumption k-Wave simulations from HU-segmented CT of 13 skulls are a valid ground truth for intracranial peak pressure
    Invoked in Section 2.2 to generate all training and test labels; any simulator error propagates directly into every metric.
  • domain assumption MR images can provide a usable skull proxy h_sk even though the paper never specifies MR acquisition, alignment, or conversion
    Table 1 reports MR inputs, but Section 2.2 only describes CT-derived h_sk; the MR pathway is asserted without protocol.
  • domain assumption 13 skulls with 300 transducer placements and 3 frequencies are representative of clinical tFUS conditions
    Section 3.1 splits into 8/2/3 skulls; the unseen-skull claim rests on three test subjects.

pith-pipeline@v1.3.0-daily-deepseek · 7692 in / 13844 out tokens · 154213 ms · 2026-08-04T19:50:38.441993+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of tFUSOperator: Operator Learning for Transcranial Focused Ultrasound Digital Twins." pith.science (2026). https://pith.science/paper/V4UTYZNS

@misc{pith2026260801839,
  author       = {Pith},
  title        = {Pith review of: tFUSOperator: Operator Learning for Transcranial Focused Ultrasound Digital Twins},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/V4UTYZNS}},
  note         = {Machine review of arXiv:2608.01839}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Transcranial focused ultrasound (tFUS) requires accurate estimation of the intracranial acoustic field, which is distorted by skull-induced aberrations. Numerical solvers are accurate but computationally expensive for digital twins, where the field must be re-estimated repeatedly as treatment conditions change. Existing deep-learning surrogates are fast but typically use voxel-to-voxel regression on a fixed grid, with no mechanism reflecting how acoustic energy propagates through the skull. We instead cast tFUS simulation as an operator learning problem and propose tFUSOperator, a coordinate-aware neural operator that maps the free-field pressure, skull anatomy, and treatment parameters to the intracranial field within a shared physical coordinate frame. To our knowledge, this is the first operator-based formulation of tFUS field prediction. On both seen and unseen skulls, the model localizes the acoustic focus accurately-reaching about 90% and 72% Dice, respectively-and it performs nearly as well from magnetic resonance (MR) as from computed tomography (CT) input while running $5.6 \times 10^4$ times faster than numerical simulation. These results suggest a fast, radiation-free route to safe and practical digital twins for patient-specific tFUS treatment. The code is available at: https://github.com/CMME-Lab/tFUSOperator.git.

Figures

Figures reproduced from arXiv: 2608.01839 by Haris Ghafoor, Kyungho Yoon, Minjee Seo, Minju Seol, Seonaeng Cho.

Figure 1
Figure 1. Figure 1: Overview of tFUSOperator. The model maps the free-field, skull volume, phys￾ical coordinates, and transducer parameters to the intracranial pressure field. Conditioning Module The conditioning module embeds treatment parame￾ters into a global condition vector for the latent processor. The module accepts an arbitrary set of scalar or vector parameters—such as transducer placement, geometry, or operating fre… view at source ↗
Figure 2
Figure 2. Figure 2: Visualization of free-field input (pff), target intracranial field (pmax), and pre￾dictions from CT (pˆ CT max) and MR (pˆ MR max) input, shown on the xz -plane through the focus for both evaluation regimes across three operating frequencies. 4 Discussion For a digital twin intended to support safe and efficient tFUS treatment, the choice of skull imaging modality is not only a matter of accuracy but also … view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

23 extracted references · 6 canonical work pages · 1 internal anchor

  1. [1]

    Alkin, B., Fürst, A., Schmid, S., Gruber, L., Holzleitner, M., Brandstetter, J.: Universal physics transformers: A framework for efficiently scaling neural operators (2025).https://doi.org/10.48550/arXiv.2402.12365

    Pith/arXiv arXiv doi:10.48550/arxiv.2402.12365 2025

  2. [2]

    di Biase, L., Falato, E., Di Lazzaro, V.: Transcranial focused ultrasound (tfus) and transcranial unfocused ultrasound (tus) neuromodulation: From theoretical principles to stimulation practices. Front. Neurol.V olume 10 - 2019(2019). https://doi.org/10.3389/fneur.2019.00549

    arXiv 2019

  3. [3]

    Cammalleri, A., Croce, P., Lee, W., Yoon, K., Yoo, S.S.: Therapeutic poten- tials of localized blood–brain barrier disruption by noninvasive transcranial fo- cused ultrasound: A technical review. J. Clin. Neurophysiol.37(2) (2020).https: //doi.org/10.1097/WNP.0000000000000488

    work page doi:10.1097/wnp.0000000000000488 2020

  4. [4]

    In: Li, L., Jirsa, V., Feng, J., Deng, J., Dede’, L., An, S., Lyu, Y., Liu, X

    Cho, S., Seo, M., Shin, M., Yoon, K.: Towards digital twin of rf ablation: Real- time prediction of time-dependent thermal effects using transformer. In: Li, L., Jirsa, V., Feng, J., Deng, J., Dede’, L., An, S., Lyu, Y., Liu, X. (eds.) Digital TwinforHealthcare.pp.69–78.SpringerNatureSwitzerland,Cham(2026).https: //doi.org/10.1007/978-3-032-07694-6_7

    work page doi:10.1007/978-3-032-07694-6_7 2026

  5. [5]

    Darmani, G., Bergmann, T., Butts Pauly, K., Caskey, C., de Lecea, L., Fomenko, A., Fouragnan, E., Legon, W., Murphy, K., Nandi, T., Phipps, M., Pinton, G., Ramezanpour, H., Sallet, J., Yaakub, S., Yoo, S., Chen, R.: Non-invasive transcra- nial ultrasound stimulation for neuromodulation. Clin. Neurophysiol.135, 51–73 (2022).https://doi.org/10.1016/j.clinph...

    doi:10.1016/j.clinph.2021.12.010 2022

  6. [6]

    Haroon, J., Aboody, K., Flores, L., McDonald, M., Mahdavi, K., Zielinski, M., Jor- dan, K., Rindner, E., Surya, J., Venkatraman, V., Go-Stevens, V., Ngai, G., Lara, J., Hyde, C., Schafer, S., Schafer, M., Bystritsky, A., Nardi, I., Kuhn, T., Ross, D., Jordan, S.: Use of transcranial low-intensity focused ultrasound for targeted delivery of stem cell-deriv...

    work page doi:10.1038/s41598-023-44785-1 2023

  7. [7]

    Jang, M., Choi, M., Jeong, I., Yoo, S.S., Yoon, K., Noh, G.: Deep learning-based real-time estimation of transcranial focused ultrasound acoustic field. Eng. Appl. Artif. Intell.156, 111157 (2025).https://doi.org/10.1016/j.engappai.2025. 111157

    doi:10.1016/j.engappai.2025 2025

  8. [8]

    Kendall, A., Gal, Y., Cipolla, R.: Multi-task learning using uncertainty to weigh losses for scene geometry and semantics (2018).https://doi.org/10.48550/ arXiv.1705.07115

    Pith/arXiv arXiv doi:10.48550/arxiv.1705.07115 2018

  9. [9]

    Kovachki, N., Li, Z., Liu, B., Azizzadenesheli, K., Bhattacharya, K., Stuart, A., Anandkumar, A.: Neural operator: Learning maps between function spaces with applications to pdes. J. Mach. Learn. Res.24(89), 1–97 (2023).https://doi.org/ 10.48550/arXiv.2108.08481 10 M. Seo et al

    Pith/arXiv arXiv doi:10.48550/arxiv.2108.08481 2023

  10. [10]

    Kyriakou, A., Neufeld, E., Werner, B., Paulides, M.M., Szekely, G., Kuster, N.: A review of numerical and experimental compensation techniques for skull-induced phase aberrations in transcranial focused ultrasound. Int. J. Hyperth.30(1), 36–46 (2014).https://doi.org/10.3109/02656736.2013.861519

    arXiv 2014

  11. [11]

    Lee, W., Kim, H.C., Jung, Y., Chung, Y.A., Song, I.U., Lee, J.H., Yoo, S.S.: Tran- scranial focused ultrasound stimulation of human primary visual cortex. Sci. Rep. 6(1), 34026 (Sep 2016).https://doi.org/10.1038/srep34026

    work page doi:10.1038/srep34026 2016

  12. [12]

    Li, Z., Meidani, K., Farimani, A.B.: Transformer for partial differential equations’ operator learning (2023).https://doi.org/10.48550/arXiv.2205.13671

    Pith/arXiv arXiv doi:10.48550/arxiv.2205.13671 2023

  13. [13]

    Naftchi-Ardebili, K., Singh, K., Popelka, G.R., Pauly, K.B.: A deep-learning model for one-shot transcranial ultrasound simulation and phase aberration correction. Med. Phys.53(1), e70259 (2026).https://doi.org/10.1002/mp.70259

    doi:10.1002/mp.70259 2026

  14. [14]

    Peebles, W., Xie, S.: Scalable diffusion models with transformers (2023).https: //doi.org/10.48550/arXiv.2212.09748

    Pith/arXiv arXiv doi:10.48550/arxiv.2212.09748 2023

  15. [15]

    Seo, M., Shin, M., Noh, G., Yoo, S.S., Yoon, K.: Multi-modal networks for real-time monitoring of intracranial acoustic field during transcranial focused ultrasound therapy. Comput. Methods Programs Biomed.257, 108458 (2024).https://doi. org/10.1016/j.cmpb.2024.108458

    arXiv 2024

  16. [16]

    In: Li, L., Jirsa, V., Feng, J., Deng, J., Dede’, L., An, S., Lyu, Y., Liu, X

    Seo, M., Shin, M., Noh, G., Yoo, S.S., Yoon, K.: Acoustic simulation with deep learning for low-intensity transcranial focused ultrasound digital twins. In: Li, L., Jirsa, V., Feng, J., Deng, J., Dede’, L., An, S., Lyu, Y., Liu, X. (eds.) Digital TwinforHealthcare.pp.58–68.SpringerNatureSwitzerland,Cham(2026).https: //doi.org/10.1007/978-3-032-07694-6_6

    doi:10.1007/978-3-032-07694-6_6 2026

  17. [17]

    Shin, M., Seo, M., Yoo, S.S., Yoon, K.: tfusformer: Physics-guided super-resolution transformer for simulation of transcranial focused ultrasound propagation in brain stimulation. IEEE J. Biomed. Health Inform.28(7), 4024–4035 (2024).https: //doi.org/10.1109/JBHI.2024.3389708

    arXiv 2024

  18. [18]

    Srivastav, V., Puel, J., Vappou, J., Houten, E.V., Cabras, P., Padoy, N.: A skull- adaptive framework for ai-based 3d transcranial focused ultrasound simulation (2025).https://doi.org/10.48550/arXiv.2505.12998

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2505.12998 2025

  19. [19]

    Treeby, B.E., Cox, B.T.: k-Wave: MATLAB toolbox for the simulation and re- construction of photoacoustic wave fields. J. Biomed. Opt.15(2), 021314 (2010). https://doi.org/10.1117/1.3360308

    doi:10.1117/1.3360308 2010

  20. [20]

    Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez,A.N.,Kaiser, L., Polosukhin, I.: Attention is all you need (2023).https://doi.org/10.48550/ arXiv.1706.03762

    Pith/arXiv arXiv doi:10.48550/arxiv.1706.03762 2023

  21. [21]

    Wang, T., Wang, C.: Latent neural operator for solving forward and inverse pde problems (2024).https://doi.org/10.48550/arXiv.2406.03923

    Pith/arXiv arXiv doi:10.48550/arxiv.2406.03923 2024

  22. [22]

    IEEE Trans

    White, J., Clement, G., Hynynen, K.: Transcranial ultrasound focus reconstruc- tion with phase and amplitude correction. IEEE Trans. Ultrason. Ferroelectr. Freq. Control52(9), 1518–1522 (2005).https://doi.org/10.1109/TUFFC.2005. 1516024

    work page doi:10.1109/tuffc.2005 2005

  23. [23]

    Yaakub, S.N., Eraifej, J., Bault, N., Lojkiewiez, M., Bellec, E., Roberts, J., Philip, N.S., Divanbeighi Zand, A.P., Green, A.L., Rushworth, M.F.S., Fouragnan, E.F.: Non-invasive ultrasonic neuromodulation of the human nucleus accumbens impacts reward sensitivity. Nat. Commun.16(1), 10192 (Nov 2025).https://doi.org/10. 1038/s41467-025-65080-9

    2025